An event compliance evaluation method and system based on artificial intelligence

Abstract

The application discloses an event compliance evaluation method and system based on artificial intelligence, which extracts multiple event keywords from an event text to be evaluated for compliance, matches corresponding compliance evaluation system texts and compliance keywords, calculates the text similarity of each event keyword and compliance keyword, thereby determining a matched keyword group, compares the semantic analysis of the event keyword in each matched keyword group and the corresponding matched compliance keyword, records the event to be evaluated for compliance as a compliance risk event if the semantics are inconsistent, adds and calculates the weight order bit number of the compliance keyword in the compliance risk event, inputs the weight order total value of the compliance keyword in the compliance risk event and the historical occurrence frequency into a pre-trained risk division model, and outputs the risk degree of the compliance risk event, thereby improving the accuracy of event compliance evaluation and the accuracy of risk level division.

Description

Technical Field

[0001] This invention relates to the field of event compliance evaluation technology, and in particular to an event compliance evaluation method and system based on artificial intelligence. Background Technology

[0002] Currently, in traditional event compliance evaluation methods, the compliance evaluation of an event is often based on the company's internal system and past experience. If the event does not comply with the company's internal system, it is considered to have a compliance risk, and the compliance risk is classified into levels based on the impact of the historical occurrence of such compliance events.

[0003] However, traditional event compliance assessment methods fail to consider the mismatch between internal company policies and actual events. In other words, as a company develops, new events often emerge faster than existing policies, potentially leaving existing policies unregulated. Furthermore, risk level classification relies solely on the degree of impact, resulting in a very simplistic approach and inaccurate risk grading. Consequently, current methods for event compliance assessment and risk level classification are generally inaccurate. Summary of the Invention

[0004] This invention provides an artificial intelligence-based method and system for evaluating event compliance, which solves the technical problems of poor accuracy in event compliance evaluation and risk level classification.

[0005] In view of this, the first aspect of the present invention provides an event compliance evaluation method based on artificial intelligence, comprising the following steps:

[0006] Obtain the compliant event text to be evaluated, perform word segmentation on the compliant event text to be evaluated, extract multiple event keywords based on the word segmentation results, and construct an event keyword library;

[0007] Based on the event type of the event text to be evaluated for compliance, match the corresponding compliance evaluation system text and compliance keywords in the preset compliance evaluation rule base;

[0008] Calculate the first text similarity between the event text to be evaluated and the compliance evaluation system text, and determine whether the first text similarity is greater than a preset first similarity threshold. If the first text similarity is greater than the preset first similarity threshold, proceed to the next step.

[0009] The compliance keywords of the compliance evaluation system text are sorted according to a preset weight. The second text similarity is calculated between each event keyword in the event keyword library and the sorted compliance keywords. It is determined whether the second text similarity is greater than the preset second similarity threshold. If the second text similarity is greater than the preset second similarity threshold, the event keyword and the corresponding compliance keyword are determined to be a matching keyword group.

[0010] Semantic analysis is performed on the event keywords and their corresponding compliant keywords in each of the matched keyword groups to obtain the semantic analysis results;

[0011] The semantic analysis results of the event keywords and their corresponding compliance keywords are compared. If the semantic comparison is inconsistent, the corresponding compliance events to be evaluated are recorded as compliance risk events.

[0012] Obtain the weight ranking positions of compliance keywords with inconsistent semantic comparison in the aforementioned compliance risk events, and sum them to obtain the total weight ranking value;

[0013] The total weight ranking of compliance keywords for the compliance risk event and its historical occurrence frequency are input into a pre-trained risk classification model, which outputs the risk level of the compliance risk event. The pre-trained risk classification model is based on a convolutional neural network and is trained with the total weight ranking of keywords for historical compliance risk events and their occurrence frequency as input, and the risk level of historical compliance risk events as output.

[0014] Preferably, the method further includes:

[0015] Extract company policy keywords from the company policy text, and construct a compliance evaluation rule base based on the company policy text and its corresponding company policy keywords;

[0016] The company policy texts are classified to obtain different types of company policy texts. Based on web crawling technology, national legal texts of the same type as the company policy texts are obtained from the public network.

[0017] A third-party text similarity calculation is performed on the national legal text and the company policy text. It is determined whether the third-party text similarity between the national legal text and the company policy text is less than a preset third-party similarity threshold. If the third-party text similarity between the national legal text and the company policy text is not less than the preset third-party similarity threshold, the national legal text is removed. If the third-party text similarity between the national legal text and the company policy text is less than the preset third-party similarity threshold, the legal keywords of the national legal text are extracted, and the national legal text and its corresponding legal keywords are stored in the compliance evaluation rule base.

[0018] Preferably, the method further includes:

[0019] The compliance evaluation system text in the preset compliance evaluation rule base is segmented into words to obtain the segmentation results;

[0020] Based on the thesaurus, multiple compliance keywords are extracted from the word segmentation results corresponding to the compliance evaluation system text;

[0021] The compliant keywords are subjected to part-of-speech analysis, and corresponding weights are assigned to the compliant keywords according to the part-of-speech type. The part-of-speech types include restrictive words, numerical words, nouns, verbs and adjectives. The weights of the part-of-speech types are ordered from largest to smallest as follows: restrictive words, numerical words, nouns, verbs and adjectives.

[0022] Preferably, the method further includes:

[0023] Compliance risk event frequency time series data is generated based on the risk level of the aforementioned compliance risk events and their corresponding historical frequency; wherein, the risk level includes minor, moderate, significant, and major.

[0024] Based on the Prophet time series prediction algorithm, a Prophet time series prediction model is constructed according to the time series data of the frequency of compliance risk events. The frequency of compliance risk events at future times is predicted by the Prophet time series prediction model to obtain the predicted value of the frequency of compliance risk events.

[0025] The predicted frequency of the compliance risk event is compared with a preset frequency threshold. If the predicted frequency of the compliance risk event exceeds the preset frequency threshold, the event compliance evaluation frequency of the corresponding compliance risk event is increased.

[0026] Secondly, the present invention also provides an artificial intelligence-based event compliance evaluation system, comprising:

[0027] The first keyword extraction module is used to obtain the compliant event text to be evaluated, perform word segmentation on the compliant event text to be evaluated, extract multiple event keywords based on the word segmentation results, and construct an event keyword library;

[0028] The keyword matching module is used to match the corresponding compliance evaluation system text and compliance keywords in the compliance evaluation system text in the preset compliance evaluation rule base according to the event type of the event text to be evaluated.

[0029] The first similarity calculation module is used to calculate the first text similarity between the event text to be evaluated and the compliance evaluation system text, and to determine whether the first text similarity is greater than a preset first similarity threshold. If the first text similarity is greater than the preset first similarity threshold, the second similarity calculation module is executed.

[0030] The second similarity calculation module is used to sort the compliance keywords of the compliance evaluation system text according to a preset weight, calculate the second text similarity between each event keyword in the event keyword library and the sorted compliance keywords, and determine whether the second text similarity is greater than a preset second similarity threshold. If the second text similarity is greater than the preset second similarity threshold, then the event keyword and the corresponding compliance keyword are determined to be a matching keyword group.

[0031] The semantic analysis module is used to perform semantic analysis on the event keywords and their corresponding compliant keywords in each of the matched keyword groups, and obtain the semantic analysis results.

[0032] The semantic comparison module is used to compare the semantic analysis results of event keywords with their corresponding compliance keywords. If the semantic comparison is inconsistent, the corresponding compliance event to be evaluated is recorded as a compliance risk event.

[0033] The weighted summation module is used to obtain the weight ranking positions of compliance keywords with inconsistent semantic comparisons in the compliance risk event and to sum them to obtain the total weight ranking value.

[0034] The risk classification module is used to input the total weight ranking of compliance keywords of the compliance risk event and the historical occurrence frequency into a pre-trained risk classification model, and output the risk level of the compliance risk event; wherein, the pre-trained risk classification model is based on a convolutional neural network, and is trained with the total weight ranking of keywords of historical compliance risk events and the occurrence frequency as input, and the risk level of historical compliance risk events as output.

[0035] Preferably, the system further includes:

[0036] The second keyword extraction module is used to extract company policy keywords from the company policy text and construct a compliance evaluation rule base based on the company policy text and its corresponding company policy keywords.

[0037] The text classification module is used to classify the company policy texts to obtain different types of company policy texts, and to obtain national legal texts of the same type as the company policy texts from the public network based on web crawling technology.

[0038] The rule base construction module is used to calculate the third-party text similarity between the national legal text and the company's institutional text, and to determine whether the third-party text similarity between the national legal text and the company's institutional text is less than a preset third-party similarity threshold. If the third-party text similarity between the national legal text and the company's institutional text is not less than the preset third-party similarity threshold, then the national legal text is removed. If the third-party text similarity between the national legal text and the company's institutional text is less than the preset third-party similarity threshold, then the legal keywords of the national legal text are extracted, and the national legal text and its corresponding legal keywords are stored in the compliance evaluation rule base.

[0039] Preferably, the system further includes:

[0040] The word segmentation module is used to perform word segmentation processing on the compliance evaluation system text in the preset compliance evaluation rule base to obtain the word segmentation result;

[0041] The third keyword extraction module is used to extract multiple compliance keywords based on the word segmentation results corresponding to the compliance evaluation system text from the thesaurus.

[0042] The part-of-speech analysis module is used to perform part-of-speech analysis on the compliant keywords and set corresponding weights for the compliant keywords according to the part-of-speech type. The part-of-speech types include restrictive words, numerical words, nouns, verbs and adjectives. The weights of the part-of-speech types are ordered from largest to smallest as follows: restrictive words, numerical words, nouns, verbs and adjectives.

[0043] Preferably, the system further includes:

[0044] The frequency and time series construction module is used to generate time series data on the frequency of compliance risk events based on the risk level of the compliance risk events and their corresponding historical frequency; wherein, the risk level includes minor, moderate, significant, and major.

[0045] The time series prediction module is used to construct a Prophet time series prediction model based on the Prophet time series prediction algorithm and the time series data of the frequency of compliance risk events. The Prophet time series prediction model is used to predict the frequency of compliance risk events at future times to obtain the predicted value of the frequency of compliance risk events.

[0046] The frequency judgment module is used to compare the predicted frequency of the compliance risk event with a preset frequency threshold. If the predicted frequency of the compliance risk event exceeds the preset frequency threshold, the event compliance evaluation frequency of the corresponding compliance risk event is increased.

[0047] Thirdly, the present invention also provides an electronic device, comprising:

[0048] At least one processor; and,

[0049] A memory communicatively connected to the at least one processor; wherein,

[0050] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described above.

[0051] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0052] As can be seen from the above technical solutions, the present invention has the following advantages:

[0053] This invention performs word segmentation on the event text to be evaluated for compliance, extracts multiple event keywords based on the segmentation results, and matches the corresponding compliance evaluation system text and compliance keywords in a preset compliance evaluation rule base according to the event type of the event text. It calculates a first text similarity between the event text and the compliance evaluation system text. If the first text similarity is greater than a preset first similarity threshold, the compliance keywords are sorted according to preset weights. A second text similarity is calculated between each event keyword and the sorted compliance keywords. The second text similarity is then determined to be greater than a preset second similarity threshold, thereby identifying matching keyword groups. Each matching keyword group is then further processed. Semantic analysis is performed on the event keywords in the keyword group and their corresponding matching compliance keywords to obtain semantic analysis results. The semantic analysis results of the event keywords and their corresponding matching compliance keywords are compared. If the semantic comparison is inconsistent, the corresponding event to be evaluated for compliance is recorded as a compliance risk event. The weight ranking positions of the compliance keywords with inconsistent semantic comparisons in the compliance risk event are summed to obtain the total weight ranking value. The total weight ranking value of the compliance keywords of the compliance risk event and the historical occurrence frequency are input into a pre-trained risk classification model to output the risk level of the compliance risk event, thereby improving the accuracy of event compliance evaluation and risk level classification. Attached Figure Description

[0054] Figure 1 A flowchart illustrating an artificial intelligence-based event compliance evaluation method provided in this embodiment of the invention;

[0055] Figure 2 This is a schematic diagram of the structure of an event compliance evaluation system based on artificial intelligence, provided as an embodiment of the present invention. Detailed Implementation

[0056] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] For easier understanding, please refer to Figure 1 The present invention provides an event compliance evaluation method based on artificial intelligence, comprising the following steps:

[0058] 101. Obtain the compliant event text to be evaluated, perform word segmentation on the compliant event text to be evaluated, extract multiple event keywords based on the word segmentation results, and construct an event keyword library.

[0059] It should be noted that when obtaining the text of the compliant event to be evaluated, the compliant event to be evaluated can be identified first. The event is recorded in the event operating system and can exist in various forms.

[0060] One method for word segmentation is to pre-build a thesaurus, then use the thesaurus for word segmentation, and extract multiple event keywords to build an event keyword library.

[0061] 102. Based on the event type of the compliance event text to be evaluated, match the corresponding compliance evaluation system text and compliance keywords in the pre-set compliance evaluation rule base.

[0062] The preset compliance evaluation rule base includes an event type index of the event text to be evaluated for compliance, as well as the corresponding compliance evaluation system text and compliance keywords in the compliance evaluation system text. The event types include human resources type, cost management type, contract type and safety type.

[0063] In one example, the LDA topic model is used to extract the topic of the compliant event text to be evaluated, and the event type of the compliant event text to be evaluated is determined based on the topic.

[0064] 103. Calculate the first text similarity between the event text and the compliance evaluation system text to be evaluated for compliance, and determine whether the first text similarity is greater than the preset first similarity threshold. If the first text similarity is greater than the preset first similarity threshold, proceed to the next step.

[0065] One approach is to map the text of the compliant event to be evaluated and the text of the compliance evaluation system into a vector space to form a numerical vector, and then use the Euclidean distance method to calculate the text similarity.

[0066] 104. Sort the compliance keywords of the compliance evaluation system text according to the preset weights. Calculate the second text similarity between each event keyword in the event keyword library and the sorted compliance keywords. Determine whether the second text similarity is greater than the preset second similarity threshold. If the second text similarity is greater than the preset second similarity threshold, then the event keyword and the corresponding compliance keyword are determined to be a matching keyword group.

[0067] 105. Perform semantic analysis on the event keywords and their corresponding compliant keywords in each matching keyword group to obtain the semantic analysis results.

[0068] One approach is to pre-train a semantic analysis model to perform semantic analysis on the event keywords and their corresponding compliance keywords in each matching keyword group.

[0069] 106. Compare the semantic analysis results of the event keywords and their corresponding compliance keywords. If the semantic comparison is inconsistent, the corresponding compliance event to be evaluated shall be recorded as a compliance risk event.

[0070] Understandably, if the textual similarity between event keywords and compliance keywords is high, but the semantic analysis results of the two are not the same, it indicates that there is a partial match. For example, if the event keyword is "exceeds" and the compliance keyword is "not exceeding", then the two are partially matched, and the corresponding compliance event to be evaluated is recorded as a compliance risk event.

[0071] 107. Obtain the weight ranking positions of compliance keywords with inconsistent semantic comparisons in compliance risk events, and sum them to obtain the total weight ranking value.

[0072] It should be noted that, since the compliance keywords were sorted according to their weights as mentioned above, if there are multiple compliance keywords with inconsistent semantic comparisons, the weight ranking positions of the compliance keywords with inconsistent semantic comparisons in the compliance risk event will be summed. If the compliance risk is higher, the weight ranking position will be higher, and the total weight ranking value obtained after summing will also be higher.

[0073] 108. Input the total weight ranking of compliance keywords of compliance risk events and their historical occurrence frequency into a pre-trained risk classification model, and output the risk level of compliance risk events; wherein, the pre-trained risk classification model is based on a convolutional neural network, and is trained with the total weight ranking of keywords of historical compliance risk events and their occurrence frequency as input, and the risk level of historical compliance risk events as output.

[0074] In one example, the affected population, scope of impact, and amount of loss caused by a compliance risk event can also be obtained. These factors, along with the total weighted value and historical frequency of the compliance risk event, are then input into the risk classification model.

[0075] Among them, the risk level of compliance risk events can be pre-classified into multiple levels, and the risk level of compliance risk events is marked during the construction of the risk classification model.

[0076] It should be noted that this embodiment provides an artificial intelligence-based event compliance evaluation method. The method involves segmenting the event text to be evaluated for compliance, extracting multiple event keywords based on the segmentation results, and matching the corresponding compliance evaluation system text and compliance keywords in a preset compliance evaluation rule base according to the event type of the event text. A first text similarity is calculated between the event text and the compliance evaluation system text. If the first text similarity is greater than a preset first similarity threshold, the compliance keywords are sorted according to preset weights. A second text similarity is calculated between each event keyword and the sorted compliance keywords. The method then determines whether the second text similarity is greater than a preset second similarity threshold, thereby confirming a match. The system identifies keyword groups and performs semantic analysis on the event keywords and their corresponding compliance keywords within each matching keyword group. The semantic analysis results are then compared with the results of the event keywords and their corresponding compliance keywords. If the semantic comparison is inconsistent, the corresponding event to be evaluated as a compliance risk event is recorded. The weight ranking positions of the compliance keywords with inconsistent semantic comparisons in the compliance risk event are summed to obtain the total weight ranking value. The total weight ranking value of the compliance keywords in the compliance risk event, along with their historical occurrence frequency, is input into a pre-trained risk classification model to output the risk level of the compliance risk event, thereby improving the accuracy of event compliance evaluation and risk level classification.

[0077] In one specific embodiment, the method further includes:

[0078] 201. Extract keywords from the company's policy documents and construct a compliance evaluation rule base based on the policy documents and their corresponding keywords.

[0079] 202. Classify the company policy documents to obtain different types of company policy documents, and use web crawling technology to obtain national legal documents of the same type as the company policy documents from the public network.

[0080] 203. Calculate the third-party text similarity between the national legal text and the company's institutional text, and determine whether the third-party text similarity between the national legal text and the company's institutional text is less than a preset third-party similarity threshold. If the third-party text similarity between the national legal text and the company's institutional text is not less than the preset third-party similarity threshold, then the national legal text is removed. If the third-party text similarity between the national legal text and the company's institutional text is less than the preset third-party similarity threshold, then the legal keywords of the national legal text are extracted, and the national legal text and its corresponding legal keywords are stored in the compliance evaluation rule base.

[0081] Specifically, if the textual similarity between a national legal text and a company policy text is not less than the preset second textual similarity, it indicates that the national legal text already has relevant institutional constraints at the company level, and therefore the national legal text is deleted. Otherwise, it is determined that the national legal text does not yet have institutional constraints at the company level, and therefore the national legal text is retained. Ultimately, all retained texts form the compliance evaluation rule base.

[0082] It is understandable that improving the compliance evaluation rule base using national laws and regulations can enable the compliance evaluation rule base to adapt to as many different types of events to be evaluated as possible.

[0083] In one specific embodiment, the method further includes:

[0084] 301. Perform word segmentation on the compliance evaluation system text in the preset compliance evaluation rule base to obtain the word segmentation results.

[0085] 302. Extract multiple compliance keywords from the word segmentation results corresponding to the compliance evaluation system text based on the thesaurus.

[0086] 303. Perform part-of-speech analysis on compliant keywords and assign corresponding weights to compliant keywords according to their part-of-speech types. The part-of-speech types include restrictive words, numerical words, nouns, verbs, and adjectives. The weights of the part-of-speech types are ordered from largest to smallest as follows: restrictive words, numerical words, nouns, verbs, and adjectives.

[0087] Among them, restrictive words are words used to limit the size range of an event, such as "greater than", "less than", and "not exceeding", while numerical words are specific numerical values. Restrictive words and numerical words better reflect whether an event is compliant, so they are given higher weight.

[0088] In one specific embodiment, the method further includes:

[0089] 401. Compliance risk event frequency time series data are generated based on the risk level of compliance risk events and their corresponding historical frequency; where risk level includes minor, general, major and significant.

[0090] Furthermore, it can also plot a time-series curve of the frequency of compliance risk events in a certain historical period based on the time-series data of the frequency of compliance risk events.

[0091] 402. Based on the Prophet time series prediction algorithm, construct a Prophet time series prediction model according to the time series data of the frequency of compliance risk events, and use the Prophet time series prediction model to predict the frequency of compliance risk events at future times to obtain the predicted value of the frequency of compliance risk events.

[0092] 403. Compare the predicted frequency of compliance risk events with the preset frequency threshold. If the predicted frequency of compliance risk events exceeds the preset frequency threshold, increase the event compliance evaluation frequency of the corresponding compliance risk event.

[0093] It should be noted that the preset frequency threshold can be set based on the frequency curve of compliance risk events in the history of the same level unit for a certain period. By comparing the longitudinal time and the horizontal comparison of the same level unit, the expected frequency and target frequency of various compliance risk events can be obtained. Based on the difference between the two, monitoring and evaluation can be strengthened to identify problems in a timely manner, make rectifications and narrow the gap, and play a guiding role in the future.

[0094] The following is a calculation example of an event compliance evaluation method based on artificial intelligence provided by the present invention.

[0095] 1) The event to be evaluated is that the final settlement amount of a certain project exceeds the budget amount. Extract the event text T1 = "The final settlement amount of a certain project exceeds the budget amount", and the event keywords A1 = "a certain project", A2 = "final settlement", A3 = "exceeds", and A4 = "budget".

[0096] 2) Determine that the event type of the compliance event to be evaluated is the cost management type. The company already has a relevant policy document T2. Extract the keywords B1, B2, B3, B4, B5, B6, B7, B8, and B9 from the relevant policy document T2, and sort the keywords B1, B2, B3, B4, B5, B6, B7, B8, and B9 in order according to the weight settings.

[0097] 3) Calculate the similarity between event text T1 and related policy text T2. If the similarity is greater than the preset similarity threshold, calculate the text similarity between the event keywords A1, A2, A3, and A4 in event text T1 and the keywords B1, B2, B3, B4, B5, B6, B7, B8, and B9 in related policy text T2 in sequence, and compare the text similarity with the preset text similarity threshold. The comparison shows that A1 matches B2, A2 matches B4, A3 matches B5, and A4 matches B8. Semantic analysis is then performed on the above-mentioned matches A1 and B2, A2 and B4, A3 and B5, and A4 and B8. It is found that the semantic analysis results of A3 and B5 are inconsistent. The semantic analysis result of A3 is "exceeds", while the semantic analysis result of B5 is "does not exceed". Therefore, it is determined that event T1 has a compliance risk, that is, this event is a compliance risk event.

[0098] 4) Obtain the ranking position of keyword B5 as 5, and obtain the frequency of compliance risk events in the historical period as F. Then, train the risk classification model and identify the risk level as relatively high. Among them, the risk level of compliance risk events is classified as S = slight, N = moderate, H = relatively high, and M = serious.

[0099] 5) Plot the frequency curve of monthly events T1 over the past n years, and predict the expected frequency F1 of monthly events N over the next n years;

[0100] 6) Based on the frequency curve of monthly events T1 in the past n years of the same level unit, set the target frequency of monthly events N F2 in the next n years;

[0101] 7) If F1 > F2 for the corresponding month, then increase the frequency of compliance evaluation for event T1.

[0102] The above is a detailed description of an embodiment of an event compliance evaluation method based on artificial intelligence provided by the present invention. The following is a detailed description of an embodiment of an event compliance evaluation system based on artificial intelligence provided by the present invention.

[0103] For easier understanding, please refer to Figure 2 The present invention also provides an artificial intelligence-based event compliance evaluation system, comprising:

[0104] The first keyword extraction module 100 is used to obtain the compliant event text to be evaluated, perform word segmentation on the compliant event text to be evaluated, extract multiple event keywords based on the word segmentation results, and build an event keyword library;

[0105] The keyword matching module 200 is used to match the corresponding compliance evaluation system text and compliance keywords in the compliance evaluation system text in the preset compliance evaluation rule base according to the event type of the compliance event text to be evaluated.

[0106] The first similarity calculation module 300 is used to calculate the first text similarity between the event text to be evaluated and the compliance evaluation system text, and to determine whether the first text similarity is greater than the preset first similarity threshold. If the first text similarity is greater than the preset first similarity threshold, the second similarity calculation module is executed.

[0107] The second similarity calculation module 400 is used to sort the compliance keywords of the compliance evaluation system text according to the preset weight, calculate the second text similarity between each event keyword in the event keyword library and the sorted compliance keywords, and determine whether the second text similarity is greater than the preset second similarity threshold. If the second text similarity is greater than the preset second similarity threshold, the event keyword and the corresponding compliance keyword are determined to be a matching keyword group.

[0108] The semantic analysis module 500 is used to perform semantic analysis on the event keywords and their corresponding compliant keywords in each matching keyword group to obtain semantic analysis results.

[0109] The semantic comparison module 600 is used to compare the semantic analysis results of the event keywords and their corresponding matching compliance keywords. If the semantic comparison is inconsistent, the corresponding compliance event to be evaluated is recorded as a compliance risk event.

[0110] The weighted summation module 700 is used to obtain the weight ranking position of compliance keywords with inconsistent semantic comparison in compliance risk events and to sum them to obtain the total weight ranking value.

[0111] The risk segmentation module 800 is used to input the total weight ranking of compliance keywords of compliance risk events and their historical occurrence frequency into a pre-trained risk segmentation model, and output the risk level of compliance risk events. The pre-trained risk segmentation model is based on a convolutional neural network and is trained with the total weight ranking of keywords of historical compliance risk events and their occurrence frequency as inputs and the risk level of historical compliance risk events as outputs.

[0112] In one specific embodiment, the system further includes:

[0113] The second keyword extraction module is used to extract keywords from the company's policy text and to build a compliance evaluation rule base based on the company's policy text and its corresponding keywords.

[0114] The text classification module is used to classify company policy texts to obtain different types of company policy texts. Based on web crawling technology, it obtains national legal texts of the same type as the company policy texts from the public network.

[0115] The rule base construction module is used to calculate the third-party text similarity between national legal texts and corporate policy texts. It determines whether the third-party text similarity between the national legal texts and corporate policy texts is less than a preset third-party similarity threshold. If the third-party text similarity between the national legal texts and corporate policy texts is not less than the preset third-party similarity threshold, the national legal text is removed. If the third-party text similarity between the national legal texts and corporate policy texts is less than the preset third-party similarity threshold, the legal keywords of the national legal texts are extracted, and the national legal texts and their corresponding legal keywords are stored in the compliance evaluation rule base.

[0116] In one specific embodiment, the system further includes:

[0117] The word segmentation module is used to segment the compliance evaluation system text in the preset compliance evaluation rule base and obtain the word segmentation results.

[0118] The third keyword extraction module is used to extract multiple compliance keywords based on the word segmentation results corresponding to the compliance evaluation system text from the thesaurus.

[0119] The part-of-speech (POS) analysis module is used to perform POS analysis on compliant keywords. It assigns corresponding weights to compliant keywords based on their POS types. The POS types include restrictive words, numerical words, nouns, verbs, and adjectives. The weights of the POS types are ordered from largest to smallest as follows: restrictive words, numerical words, nouns, verbs, and adjectives.

[0120] In one specific embodiment, the system further includes:

[0121] The frequency and time series construction module is used to generate time series data on the frequency of compliance risk events based on their risk level and the frequency of occurrence in their corresponding historical periods; where the risk level includes minor, moderate, significant, and major.

[0122] The time series prediction module is used to construct a Prophet time series prediction model based on the Prophet time series prediction algorithm and the time series data of the frequency of compliance risk events. The Prophet time series prediction model is used to predict the frequency of compliance risk events at future times and obtain the predicted value of the frequency of compliance risk events.

[0123] The frequency judgment module is used to compare the predicted frequency of compliance risk events with the preset frequency threshold. If the predicted frequency of compliance risk events exceeds the preset frequency threshold, the event compliance evaluation frequency of the corresponding compliance risk event will be increased.

[0124] The present invention also provides an electronic device, comprising:

[0125] At least one processor; and,

[0126] A memory that is communicatively connected to at least one processor; wherein,

[0127] The memory stores instructions that can be executed by at least one processor, such that the instructions are executed by at least one processor to enable the at least one processor to perform the method described above.

[0128] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0129] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, electronic devices, and computer-readable storage media described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0130] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0131] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0132] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0133] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for executing all or part of the steps of the methods described in the various embodiments of this invention through a computer device (which may be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0134] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An event compliance evaluation method based on artificial intelligence, characterized in that, Includes the following steps: Obtain the compliant event text to be evaluated, perform word segmentation on the compliant event text to be evaluated, extract multiple event keywords based on the word segmentation results, and construct an event keyword library; Based on the event type of the event text to be evaluated for compliance, match the corresponding compliance evaluation system text and compliance keywords in the preset compliance evaluation rule base; Calculate the first text similarity between the event text to be evaluated and the compliance evaluation system text, and determine whether the first text similarity is greater than a preset first similarity threshold. If the first text similarity is greater than the preset first similarity threshold, proceed to the next step. The compliance keywords of the compliance evaluation system text are sorted according to a preset weight. The second text similarity is calculated between each event keyword in the event keyword library and the sorted compliance keywords. It is determined whether the second text similarity is greater than the preset second similarity threshold. If the second text similarity is greater than the preset second similarity threshold, the event keyword and the corresponding compliance keyword are determined to be a matching keyword group. Semantic analysis is performed on the event keywords and their corresponding compliant keywords in each of the matched keyword groups to obtain the semantic analysis results; The semantic analysis results of the event keywords and their corresponding compliance keywords are compared. If the semantic comparison is inconsistent, the corresponding compliance events to be evaluated are recorded as compliance risk events. Obtain the weight ranking positions of compliance keywords with inconsistent semantic comparison in the aforementioned compliance risk events, and sum them to obtain the total weight ranking value; The total weight ranking of compliance keywords for the compliance risk event and its historical occurrence frequency are input into a pre-trained risk classification model, which outputs the risk level of the compliance risk event. The pre-trained risk classification model is based on a convolutional neural network and is trained with the total weight ranking of keywords for historical compliance risk events and their occurrence frequency as input, and the risk level of historical compliance risk events as output.

2. The event compliance evaluation method based on artificial intelligence according to claim 1, characterized in that, Also includes: Extract keywords from the company's policy documents and construct a compliance evaluation rule base based on the policy documents and their corresponding keywords. The company policy texts are classified to obtain different types of company policy texts. Based on web crawling technology, national legal texts of the same type as the company policy texts are obtained from the public network. A third-party text similarity calculation is performed on the national legal text and the company policy text. It is determined whether the third-party text similarity between the national legal text and the company policy text is less than a preset third-party similarity threshold. If the third-party text similarity between the national legal text and the company policy text is not less than the preset third-party similarity threshold, the national legal text is removed. If the third-party text similarity between the national legal text and the company policy text is less than the preset third-party similarity threshold, the legal keywords of the national legal text are extracted, and the national legal text and its corresponding legal keywords are stored in the compliance evaluation rule base.

3. The event compliance evaluation method based on artificial intelligence according to claim 1, characterized in that, Also includes: The compliance evaluation system text in the preset compliance evaluation rule base is segmented into words to obtain the segmentation results; Based on the thesaurus, multiple compliance keywords are extracted from the word segmentation results corresponding to the compliance evaluation system text; The compliant keywords are subjected to part-of-speech analysis, and corresponding weights are assigned to the compliant keywords according to the part-of-speech type. The part-of-speech types include restrictive words, numerical words, nouns, verbs and adjectives. The weights of the part-of-speech types are ordered from largest to smallest as follows: restrictive words, numerical words, nouns, verbs and adjectives.

4. The event compliance evaluation method based on artificial intelligence according to claim 1, characterized in that, Also includes: Compliance risk event frequency time series data is generated based on the risk level of the aforementioned compliance risk events and their corresponding historical frequency; wherein, the risk level includes minor, moderate, significant, and major. Based on the Prophet time series prediction algorithm, a Prophet time series prediction model is constructed according to the time series data of the frequency of compliance risk events. The frequency of compliance risk events at future times is predicted by the Prophet time series prediction model to obtain the predicted value of the frequency of compliance risk events. The predicted frequency of the compliance risk event is compared with a preset frequency threshold. If the predicted frequency of the compliance risk event exceeds the preset frequency threshold, the event compliance evaluation frequency of the corresponding compliance risk event is increased.

5. An event compliance evaluation system based on artificial intelligence, characterized in that, include: The first keyword extraction module is used to obtain the compliant event text to be evaluated, perform word segmentation on the compliant event text to be evaluated, extract multiple event keywords based on the word segmentation results, and construct an event keyword library; The keyword matching module is used to match the corresponding compliance evaluation system text and compliance keywords in the compliance evaluation system text in the preset compliance evaluation rule base according to the event type of the event text to be evaluated. The first similarity calculation module is used to calculate the first text similarity between the event text to be evaluated and the compliance evaluation system text, and to determine whether the first text similarity is greater than a preset first similarity threshold. If the first text similarity is greater than the preset first similarity threshold, the second similarity calculation module is executed. The second similarity calculation module is used to sort the compliance keywords of the compliance evaluation system text according to a preset weight, calculate the second text similarity between each event keyword in the event keyword library and the sorted compliance keywords, and determine whether the second text similarity is greater than a preset second similarity threshold. If the second text similarity is greater than the preset second similarity threshold, then the event keyword and the corresponding compliance keyword are determined to be a matching keyword group. The semantic analysis module is used to perform semantic analysis on the event keywords and their corresponding compliant keywords in each of the matched keyword groups, and obtain the semantic analysis results. The semantic comparison module is used to compare the semantic analysis results of event keywords with their corresponding compliance keywords. If the semantic comparison is inconsistent, the corresponding compliance event to be evaluated is recorded as a compliance risk event. The weighted summation module is used to obtain the weight ranking positions of compliance keywords with inconsistent semantic comparisons in the compliance risk event and to sum them to obtain the total weight ranking value. The risk classification module is used to input the total weight ranking of compliance keywords of the compliance risk event and the historical occurrence frequency into a pre-trained risk classification model, and output the risk level of the compliance risk event; wherein, the pre-trained risk classification model is based on a convolutional neural network, and is trained with the total weight ranking of keywords of historical compliance risk events and the occurrence frequency as input, and the risk level of historical compliance risk events as output.

6. The event compliance evaluation system based on artificial intelligence according to claim 5, characterized in that, Also includes: The second keyword extraction module is used to extract keywords from the company's policy text and construct a compliance evaluation rule base based on the company's policy text and its corresponding keywords. The text classification module is used to classify the company policy texts to obtain different types of company policy texts, and to obtain national legal texts of the same type as the company policy texts from the public network based on web crawling technology. The rule base construction module is used to calculate the third-party text similarity between the national legal text and the company's institutional text, and to determine whether the third-party text similarity between the national legal text and the company's institutional text is less than a preset third-party similarity threshold. If the third-party text similarity between the national legal text and the company's institutional text is not less than the preset third-party similarity threshold, then the national legal text is removed. If the third-party text similarity between the national legal text and the company's institutional text is less than the preset third-party similarity threshold, then the legal keywords of the national legal text are extracted, and the national legal text and its corresponding legal keywords are stored in the compliance evaluation rule base.

7. The event compliance evaluation system based on artificial intelligence according to claim 5, characterized in that, Also includes: The word segmentation module is used to perform word segmentation processing on the compliance evaluation system text in the preset compliance evaluation rule base to obtain the word segmentation result; The third keyword extraction module is used to extract multiple compliance keywords based on the word segmentation results corresponding to the compliance evaluation system text from the thesaurus. The part-of-speech analysis module is used to perform part-of-speech analysis on the compliant keywords and set corresponding weights for the compliant keywords according to the part-of-speech type. The part-of-speech types include restrictive words, numerical words, nouns, verbs and adjectives. The weights of the part-of-speech types are ordered from largest to smallest as follows: restrictive words, numerical words, nouns, verbs and adjectives.

8. The event compliance evaluation system based on artificial intelligence according to claim 5, characterized in that, Also includes: The frequency and time series construction module is used to generate time series data on the frequency of compliance risk events based on the risk level of the compliance risk events and their corresponding historical frequency; wherein, the risk level includes minor, moderate, significant, and major. The time series prediction module is used to construct a Prophet time series prediction model based on the Prophet time series prediction algorithm and the time series data of the frequency of compliance risk events. The Prophet time series prediction model is used to predict the frequency of compliance risk events at future times to obtain the predicted value of the frequency of compliance risk events. The frequency judgment module is used to compare the predicted frequency of the compliance risk event with a preset frequency threshold. If the predicted frequency of the compliance risk event exceeds the preset frequency threshold, the event compliance evaluation frequency of the corresponding compliance risk event is increased.

9. An electronic device, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method as described in any one of claims 1 to 4.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 4.

Images