Early warning content risk quality control method and system based on multi-granularity semantic analysis

Abstract

The application discloses a pre-warning content risk quality control method and system based on multi-granularity semantic analysis, and relates to the technical field of intelligent processing of pre-warning information; the method comprises the following steps: obtaining pre-warning data to be controlled and pre-warning level specification data; identifying conditional content in the pre-warning text to obtain text conditional group data; identifying a text conditional group in the text conditional group data for supporting the pre-warning level to obtain level support data; identifying supplementary description data by supplementally describing the text conditional group which does not form the level support; and controlling the level consistency between the pre-warning level and the pre-warning text to obtain a level consistency quality control result; the application can reduce the quality control deviation caused by the error combination of pre-warning elements, and improve the accuracy of pre-warning text risk quality control.

Description

Technical Field

[0001] This application relates to the field of intelligent processing technology for early warning information, and more specifically, to a method and system for risk quality control of early warning content based on multi-granularity semantic analysis. Background Technology

[0002] In the context of early warning content release and risk quality control, risk quality control of early warning content is a crucial step in ensuring the accuracy and standardization of early warning information. Early warning information, such as weather warnings, typically requires providing the warning type, warning level, and the main text of the warning. The main text also includes information such as the affected area, the period of impact, changes in meteorological indicators, numerical ranges, development trends, and risk warnings. The warning level should generally correspond to the degree of risk expressed in the warning text; if the warning level and the text are inconsistent, it can easily lead to a mismatch between the risk level expressed in the warning information and the published text, affecting the reliability of the early warning content review results.

[0003] Existing methods for reviewing warning content typically employ format validation, fixed field verification, keyword matching, warning element extraction, and rule comparison. Specifically, the system identifies warning elements such as time, indicators, values, and status from the warning text, then compares these elements with the business rules or regulatory conditions corresponding to the warning level to determine if there is a discrepancy in level consistency between the warning level and the warning text. This approach can check whether the warning text contains elements related to the warning level and can also verify the correspondence between some numerical descriptions and regulatory conditions.

[0004] However, in actual published texts, in addition to the descriptions used as the basis for determining the current warning level, the warning text may also include supplementary risk descriptions such as local changes, subsequent evolution, ongoing impact, and risk warnings. Both the descriptions used as the basis for determining the current warning level and the supplementary risk descriptions may contain similar linguistic structures such as time, indicators, values, and status. Current warning content review methods, when conducting quality control for level consistency, typically focus more on whether warning elements have been identified and whether the identified local warning elements can match the normative conditions corresponding to the warning level. They struggle to determine whether different condition descriptions within the same warning text are based on the current warning level.

[0005] When multiple time expressions, indicator descriptions, or numerical descriptions coexist within the same warning text, existing warning content review methods are prone to combining time, indicator, numerical, and status information belonging to different conditions, creating judgment relationships not originally expressed in the warning text. Such incorrect combinations may lead to supplementary risk descriptions being used as the basis for warning level judgments, or may cause the textual basis truly corresponding to the warning level to be interfered with by other local descriptions. Therefore, when faced with warning texts that simultaneously contain level basis descriptions and supplementary risk descriptions, existing warning content review methods are prone to misjudging or omitting quality control results, affecting the accuracy of risk quality control conclusions for warning content.

[0006] In view of this, this application proposes a method and system for early warning content risk quality control based on multi-granularity semantic analysis to solve the above problems. Summary of the Invention

[0007] To reduce quality control bias caused by the incorrect use of supplementary risk descriptions as the basis for judging warning levels, and to improve the accuracy of risk quality control in warning texts that include both level-based descriptions and supplementary risk descriptions, this application provides the following technical solution: a risk quality control method for warning content based on multi-granularity semantic analysis, including:

[0008] Obtain the quality control early warning data and early warning level specification data, wherein the quality control early warning data includes the early warning type, early warning level, and early warning text;

[0009] Combining the warning type and the warning level, the judgment conditions corresponding to the warning type and the warning level are extracted from the warning level specification data to obtain the level judgment condition data;

[0010] The conditional content in the warning text is identified, and the conditional content used to represent the same condition is grouped to obtain the text conditional group data;

[0011] Based on the level determination condition data, identify the text condition groups in the text condition group data that support the warning level, and obtain the level support data;

[0012] Based on the level support data, supplementary descriptions are identified for text condition groups that do not form level support, and supplementary description data is obtained.

[0013] By combining the level support data, the supplementary description data, and the level determination condition data, quality control is performed on the level consistency between the warning level and the warning text to obtain the level consistency quality control result.

[0014] Furthermore, the method for obtaining the text condition group data includes:

[0015] The warning text is segmented according to the pre-defined sentence segmentation boundaries to obtain the main text sentences;

[0016] According to the pre-configured conditional content recognition rules, the main text statement is subjected to text recognition to obtain multiple conditional contents, and the content type of each conditional content is recorded. The content type is determined from the indicator object, time limit, numerical judgment content and occurrence status.

[0017] Filter the content of the multiple conditional contents that is of the type of indicator object, and extract the indicator objects from the filtered conditional contents;

[0018] Based on the extracted indicator objects, the conditional content that is in the same sentence as the indicator object or has a semantic connection relationship is grouped to form the main text conditional group;

[0019] Assign a text condition group identifier to the text condition group to form text condition group data.

[0020] Furthermore, the method for determining the existence of semantic succession relationships includes:

[0021] In adjacent text statements, read the text condition groups that have been established in the text statements that are sorted first, and read the text condition group identifier and index object corresponding to the text condition group.

[0022] Using the text condition group identifier as the processing unit, check whether the text condition group has recorded the time limit, numerical judgment content and occurrence status, and determine the unrecorded time limit, numerical judgment content or occurrence status as the content type to be supplemented.

[0023] According to the pre-configured semantic continuation expression, the conditional content in the subsequent text statements is semantically matched with the index object corresponding to the text condition group identifier to obtain the continuation matching result.

[0024] When the matching result is that the following text statement does not introduce new indicator objects, and condition content consistent with the type of content to be supplemented is identified from the following text statement, it is determined that the following text statement and the indicator object corresponding to the text condition group identifier have a semantic connection relationship.

[0025] Furthermore, the method for obtaining the level support data includes:

[0026] Extract judgment condition records from the level judgment condition data, and extract text condition group records from the text condition group data;

[0027] Using a single text condition group record and a single decision condition record as comparison objects, the condition content type of the actual record in the decision condition record is read as the condition content type to be compared.

[0028] For each condition content type involved in the comparison, the condition content in the main text condition group record and the judgment condition record is compared; for the condition content types that form a corresponding condition, a corresponding mark and a corresponding condition content type are written; for the condition content types that do not form a corresponding condition, a non-corresponding mark and a non-corresponding condition content type are written, and a condition comparison record is generated.

[0029] Based on the conditions, comparison records are generated to produce comparison results that show all corresponding records, some corresponding records, or no corresponding records.

[0030] When the comparison result between the main text condition group record and any decision condition record is completely consistent, the main text condition group record is marked as forming level support; when the comparison result between the main text condition group record and each decision condition record is partially consistent or not consistent, the main text condition group record is marked as not forming level support; the marking results, condition comparison records and comparison results are summarized to form level support data.

[0031] Furthermore, the method for obtaining supplementary descriptive data includes:

[0032] Read the text condition groups that do not form level support, along with the corresponding condition comparison records and comparison results, from the level support data.

[0033] From the condition comparison records and the comparison results, filter out the condition comparison records that correspond to some of the comparison results to obtain the condition comparison records to be identified;

[0034] According to the pre-configured supplementary description identification rules, the corresponding and non-corresponding condition content types in the comparison record of the conditions to be identified are judged;

[0035] When the corresponding condition content type includes an indicator object, and also includes at least one of time limit, numerical judgment content, or occurrence status, and the uncorresponding condition content type includes at least one of time limit, numerical judgment content, or occurrence status of the actual record of the judgment condition, the text condition group forming the comparison record of the condition to be identified will be identified as a supplementary description text condition group.

[0036] Write the supplementary description text condition group, the judgment condition for forming the comparison record of the condition to be identified, the comparison record of the condition to be identified, the content type of the corresponding condition, and the content type of the uncorresponding condition into the supplementary description data.

[0037] Furthermore, the method for obtaining the level consistency quality control results includes:

[0038] Read the tagging results, condition comparison records, and comparison results from the level support data, and read the supplementary description text condition groups from the supplementary description data;

[0039] Filter the text condition groups marked as forming level support from the marked results. If the filtering results are not empty, generate level basis forming tags and write the filtered text condition groups into the level consistency quality control results.

[0040] If the filtering result is empty, the judgment condition is read from the level judgment condition data, and based on the condition comparison record and the comparison result, it is judged whether each judgment condition forms a complete set of corresponding comparison results with any text condition group. The judgment condition that does not form a complete set of corresponding comparison results is determined as a level basis insufficient judgment condition, a level basis insufficient mark is generated, and the level basis insufficient judgment condition is written into the level consistency quality control result.

[0041] If the supplementary description data includes at least one supplementary description text condition group, a supplementary description prompt mark is generated, and the supplementary description text condition group is written into the level consistency quality control result.

[0042] Furthermore, the method for generating supplementary description prompts includes:

[0043] Read the supplementary description text condition group, condition comparison record, judgment condition for forming the condition comparison record, corresponding condition content type and non-corresponding condition content type from the supplementary description data;

[0044] Write the supplementary description text condition group, the judgment condition, the condition comparison record, the corresponding condition content type and the non-corresponding condition content type into the supplementary description text condition group field, the judgment condition field, the condition comparison record field, the corresponding type field and the non-corresponding type field respectively to obtain the quality control prompt content;

[0045] Generate supplementary description prompts and write the quality control prompts into the level consistency quality control results.

[0046] Furthermore, the method for comparing the numerical judgment content in the text condition group record and the judgment condition record includes:

[0047] When the condition content type involved in the comparison is numerical judgment content, the numerical limitation direction, limitation value, and unit of measurement are extracted from the numerical judgment content in the main text condition group record and the judgment condition record, respectively.

[0048] When the unit of measurement in the main text condition group record is inconsistent with the unit of measurement in the judgment condition record, the units of the limited values ​​in the main text condition group record and the limited values ​​in the judgment condition record are unified according to the pre-configured unit conversion rules.

[0049] If the unit of measurement in the main text condition group record is consistent with the unit of measurement in the judgment condition record, or after the unit is unified, the corresponding judgment is made on the numerical limitation direction and limitation value in the main text condition group record and the numerical limitation direction and limitation value in the judgment condition record according to the preset numerical judgment correspondence rules, and the numerical judgment content judgment result is obtained.

[0050] When the result of the numerical determination is that a correspondence is formed, a corresponding mark is written into the condition comparison record; when the result of the numerical determination is that a correspondence is not formed, a non-corresponding mark is written into the condition comparison record.

[0051] Furthermore, the method for obtaining the level determination condition data includes:

[0052] Search the warning level specification data for a level determination statement that matches the warning type and the warning level.

[0053] The level determination statement is subjected to text normalization processing, and the normalized level determination statement is subjected to condition content extraction to obtain the condition content extraction result. The content type of the condition content extraction result is determined from the indicator object, time limit, numerical determination content and occurrence status.

[0054] Identify the common-factual expressions and alternative-factual expressions in the level judgment statement, and combine or segment the condition content extraction results based on the common-factual expressions and alternative-factual expressions to obtain the judgment conditions;

[0055] Assign a judgment condition identifier to the judgment condition, organize the judgment condition identifier, the indicator object, time limit, numerical judgment content and occurrence status in the judgment condition into a judgment condition record, and summarize the judgment condition record to obtain the level judgment condition data.

[0056] A content risk control system based on multi-granularity semantic analysis includes:

[0057] The data acquisition module is used to acquire the quality control early warning data and the early warning level specification data. The quality control early warning data includes the early warning type, early warning level and early warning text.

[0058] The level determination condition forming module is used to combine the warning type and the warning level, extract the determination conditions corresponding to the warning type and the warning level from the warning level specification data, and obtain the level determination condition data.

[0059] The text condition group forming module is used to identify the condition content in the warning text and group the condition content that represents the same condition to obtain text condition group data.

[0060] The level support data forming module is used to identify the text condition groups in the text condition group data that support the warning level based on the level determination condition data, and to obtain level support data;

[0061] The supplementary description data forming module is used to identify supplementary descriptions for text condition groups that have not formed level support based on the level support data, and obtain supplementary description data.

[0062] The quality control result generation module is used to combine the level support data, the supplementary description data, and the level judgment condition data to perform quality control on the level consistency between the warning level and the warning text, and obtain the level consistency quality control result.

[0063] Compared with existing technologies, the technical effects and advantages of the early warning content risk quality control method and system based on multi-granularity semantic analysis proposed in this application are as follows:

[0064] This application focuses on quality control of the consistency between warning levels and warning text. It extracts the judgment conditions corresponding to warning types and levels from the warning level specification data and groups the condition content used to represent the same condition in the warning text. This allows for subsequent comparisons between the level judgment criteria on the specification side and the condition descriptions on the text side under the same processing caliber. Compared to existing technologies that directly match rules based on scattered time, indicators, values, and states, this application reduces element mismatches caused by scattered identification of warning elements and improves the input consistency of level consistency quality control.

[0065] When the warning text contains multiple time expressions, multiple indicator descriptions, or multiple numerical descriptions, the text condition group serves as the level support judgment object and participates in the identification of the level judgment conditions, thus limiting the level judgment basis to the same condition description range. The time, indicator, value, and status in different condition descriptions will not be directly concatenated into a judgment relationship that was not originally expressed in the warning text, thereby reducing quality control misjudgments or omissions caused by cross-condition combination.

[0066] For text condition groups that lack level-based support, the system continues to identify supplementary descriptions and uses these supplementary descriptions, along with level-based support data and level-determination condition data, to generate level consistency quality control results. This process distinguishes between text condition descriptions used as the basis for current warning level judgments and supplementary risk descriptions such as local changes, subsequent evolution, ongoing impacts, and risk warnings, preventing supplementary risk descriptions from being directly used as the basis for warning level judgments.

[0067] The continuous processing chain, formed by extracting level determination conditions, grouping textual conditions, identifying level support, and recognizing supplementary descriptions, ensures that risk quality control of early warning content no longer relies on partial element matching or single rule matching. Instead, it generates level-consistent quality control results based on the relationship between regulatory basis, textual basis, and supplementary risk descriptions. This improves upon existing early warning content review methods that suffer from scattered identification, partial matching, and incorrect grouping of early warning elements, resulting in more accurate risk quality control conclusions for early warning texts that include level basis descriptions and supplementary risk descriptions. Attached Figure Description

[0068] Figure 1 This is a schematic diagram illustrating the warning content release and risk control environment in an embodiment of this application.

[0069] Figure 2 This is a flowchart of the early warning content risk quality control method based on multi-granularity semantic analysis according to an embodiment of this application;

[0070] Figure 3 This is a schematic diagram of a risk control system for early warning content based on multi-granularity semantic analysis, as described in an embodiment of this application. Detailed Implementation

[0071] The technical solutions of this application will be described in detail, clearly, and completely below with reference to the accompanying drawings of the embodiments. It should be particularly noted that the specific embodiments described below are only used to better illustrate and explain the technical solutions of this application, and are intended to enable those skilled in the art to better understand and implement this application, and should not be construed as limiting the scope of protection of this application. Without departing from the spirit and substance of this application, those skilled in the art can modify, adjust, or make equivalent substitutions based on the content disclosed in this application, and these modifications, adjustments, or equivalent substitutions should all be considered within the scope of protection of this application.

[0072] See Figure 1 As shown in the figure, this application embodiment provides an environment for issuing early warning content and controlling risk. The environment for issuing early warning content and controlling risk includes an early warning content issuing system 101, an early warning content risk control device 102, and an early warning specification library 103.

[0073] The early warning content publishing system 101 is used to generate or receive early warning content to be published. The early warning content includes structured fields and unstructured text. The structured fields record information such as the early warning type, early warning level, publication time, and affected area, while the unstructured text records the main body of the early warning. The publication time and affected area are stored as additional fields in the early warning content. In this embodiment, the level consistency quality control reads the early warning type, early warning level, and early warning text, but not the publication time and affected area. When the early warning content enters the pre-publication risk quality control stage, the early warning content publishing system 101 provides the early warning data to be controlled to the early warning content risk quality control device 102 through a data interface.

[0074] The early warning specification library 103 stores early warning level specification data. This data is organized according to early warning type and level, and includes the early warning type, level, and level determination statement. The early warning specification library 103 provides the early warning level specification data to the early warning content risk control device 102 via a data interface.

[0075] The early warning content risk quality control device 102 is used to receive early warning data to be quality controlled and to call the early warning level specification data. The early warning content risk quality control device 102 takes the early warning data to be quality controlled and the early warning level specification data as input, performs quality control on the consistency between the early warning level and the early warning text, and outputs the level consistency quality control result.

[0076] The early warning content publishing system 101, the early warning content risk control device 102, and the early warning specification library 103 can be deployed on the same computing device or on different computing devices that are interconnected. Data interaction between the early warning content publishing system 101, the early warning content risk control device 102, and the early warning specification library 103 can be achieved through interface calls, database access, or message transmission. This application embodiment does not limit the specific deployment form or the specific data interaction method.

[0077] exist Figure 1 In the warning content release and risk quality control environment shown, the warning data to be controlled is used to provide the warning content that needs risk quality control, the warning level specification data is used to provide the specification basis related to the warning level, and the level consistency quality control result is the output result generated by the warning content risk quality control device 102 based on the warning data to be controlled and the warning level specification data. After completing the description of the warning content release and risk quality control environment, please refer to... Figure 2 As shown, the risk control method for early warning content based on multi-granularity semantic analysis executed by the early warning content risk control device 102 includes:

[0078] S1: Obtain the data to be controlled and ...

[0079] In practice, when a warning content enters the risk control stage, the risk control device reads the warning data to be controlled from the warning content release system. The warning data to be controlled includes the warning type, warning level, and warning text, all of which correspond to the same warning content.

[0080] The risk control device for early warning content uses the early warning type and level as the basis for locating the level determination condition data in S2, and uses the early warning text as the text processing object for generating the text condition group data in S3.

[0081] The risk control device for early warning content reads the early warning level specification data from the early warning specification library. The early warning level specification data is pre-organized according to the early warning issuance specifications, business rule documents, or applicable local rules, and is managed according to the early warning type and early warning level.

[0082] The warning level specification data includes warning type, warning level, and level determination statement. The level determination statement is used to record the specification statement content of the determination requirements for the corresponding warning type and warning level, and serves as the basis for extracting the level determination condition data in S2.

[0083] S2: Combining the warning type and warning level, extract the judgment conditions corresponding to the warning type and warning level from the warning level standard data to obtain the level judgment condition data.

[0084] In practice, the risk control device for early warning content reads the early warning type and level from the early warning data to be controlled, and searches for level determination statements in the early warning level specification data that are consistent with the early warning type and level.

[0085] The risk quality control device for early warning content performs condition extraction processing on the level determination statement. Condition extraction processing identifies the indicator object, time limit, numerical determination content, and occurrence status from the level determination statement that characterize the determination requirements for the corresponding early warning level, and uses the identification results to organize the determination conditions. The indicator object characterizes the type of meteorological indicator participating in the level determination; the time limit characterizes the statistical period, forecast period, or duration period corresponding to the meteorological indicator; the numerical determination content records the level establishment requirements for the indicator object's value in the early warning level specification data, including the numerical limitation direction, the limited value, and the unit of measurement. The numerical limitation direction includes greater than, less than, greater than or equal to, and less than or equal to; the numerical determination content is compared with the numerical determination content in the text condition group in S4 to determine whether the numerical description of the same indicator object in the early warning text meets the numerical requirements for the corresponding early warning level in the early warning level specification data; the occurrence status characterizes the state requirements of the meteorological indicator having occurred, being expected to occur, or continuing to exist.

[0086] The risk control device for early warning content identifies the connection expressions used to link different identification results in the level determination statement. These connection expressions are identified according to pre-configured connection expression identification rules, which are pre-configured based on text expressions in the early warning specification statement used to indicate whether conditions are both true or only one is true. The risk control device for early warning content distinguishes between connection expressions indicating both conditions are true and connection expressions indicating only one condition is true.

[0087] When multiple identification results are connected by a linking expression indicating that they are both true, the risk control device for early warning content organizes these results into a single judgment condition. When a level judgment statement contains a linking expression indicating that one of the criteria is true, the risk control device for early warning content segments the statement using this linking expression as the boundary, and organizes the identification results within each segment. When the identification results within a segment include an indicator object, and include one or more of the following: a time limit, numerical judgment content, or occurrence state that together with the indicator object defines the judgment requirement for the corresponding early warning level, and the identification results within the segment do not need to be true together with the identification results within other segments to meet the judgment requirement for the early warning level, the risk control device for early warning content organizes the identification results within the segment into a single judgment condition.

[0088] The risk control device for early warning content assigns a judgment condition identifier to each judgment condition and organizes these conditions into a judgment condition record. The judgment condition record includes the judgment condition identifier, the indicator object, the time limit, the numerical judgment content, and the occurrence status. For time limits, numerical judgment content, or occurrence statuses in the judgment condition that are not actually recorded, the risk control device for early warning content writes a "not recorded" mark in the judgment condition record. This "not recorded" mark indicates that the type of condition content not actually recorded in the judgment condition will not participate in the comparison in S4.

[0089] The risk control device for early warning content will summarize the judgment condition records corresponding to the early warning type and level to obtain level judgment condition data. The level judgment condition data is used to compare with the text condition group data in S4 to identify the text condition groups in the text condition group data that support the early warning level.

[0090] S3: Identify the conditional content in the warning text and group the conditional content that represents the same condition to obtain the text conditional group data.

[0091] In practice, the risk control device for early warning content reads the main text of the early warning data to be controlled and segments it according to pre-defined sentence segmentation boundaries. These boundaries include periods, semicolons, line breaks, and segmentation markers in the early warning release template. Following pre-configured conditional content recognition rules, the device performs text recognition processing on each sentence, extracting the indicator object, time limit, numerical judgment content, and occurrence status.

[0092] The conditional content identification rules include rules for expressing indicator objects, rules for expressing time limits, rules for expressing numerical judgment content, and rules for expressing occurrence status. These rules are stored in an expression rule table, which includes a rule identifier, content type, matching expression, normalized field, and applicable warning type. The warning content risk control device reads the main text statements in the order of their identifiers and calls the expression rule table in the order of indicator object, time limit, numerical judgment content, and occurrence status. When continuous text in the main text statement matches a matching expression, the continuous text is used as the original text content, the normalized field corresponding to the matching expression is used as the normalized text, and the content type, original text position, and corresponding statement identifier are recorded to form a conditional content record. If the same continuous text matches more than two matching expressions, the matching expression whose applicable warning type matches the warning type in the warning data to be controlled is preferred; if multiple matching expressions still exist, the matching expression with the largest number of matching characters is used. The warning content risk control device records the content type, original text position, original text content, and normalized text for each identified conditional content.

[0093] The risk control device for early warning content groups the identified conditions according to condition grouping rules. These rules determine whether multiple conditions collectively represent the same condition. The condition grouping rules are pre-configured based on the text structure used to organize the description of the same condition in early warning release specifications, business rule documents, or applicable local rules. The text structure of the same condition includes multiple conditions in the same statement that jointly define the same indicator object, and subsequent statements in adjacent statements supplementing the previous statement with any unrecorded conditions under the same indicator object.

[0094] When one or more conditions—time limits, numerical judgments, or occurrence states—in the same statement jointly define the same indicator object and do not form another set of independent conditions, the warning content risk control device will group the indicator object and the conditions used to define the same indicator object into the same text condition group. When the same statement forms two or more sets of independent conditions around the same indicator object, and each set of conditions constitutes an independent condition around different time limits, different numerical judgments, or different occurrence states, the warning content risk control device will establish separate text condition groups.

[0095] For adjacent text statements, the risk control device for early warning content treats the text statement that appears earlier as the preceding text statement and the text statement that appears later as the following text statement. The device reads the established text condition groups in the preceding text statement one by one, and reads the corresponding text condition group identifier and indicator object. Using the text condition group identifier as the processing unit, it checks whether the text condition group has recorded time limits, numerical judgment content, and occurrence status. Any unrecorded time limits, numerical judgment content, or occurrence status is identified as content types to be supplemented. The device then performs semantic connection matching between the condition content in the following text statement and the indicator object corresponding to the text condition group identifier, according to pre-configured semantic connection expressions, to obtain the connection matching result. The semantic connection expressions are pre-configured based on text expressions used to indicate content continuation, partial supplementation, or continuous explanation in early warning release specifications, business rule documents, or applicable local rules. Semantic continuation expressions are stored in the form of a semantic continuation expression table. This table includes the continuation expression, the allowed preceding content type, the allowed supplementary content type, and a prohibited continuation marker. The early warning content risk control device reads the opening text, referential text, and connecting text from the following main text statement and compares them item by item with the continuation expressions in the semantic continuation expression table. If the following main text statement matches a continuation expression, the allowed supplementary content type recorded in the semantic continuation expression table matches the type of content to be supplemented, and the following main text statement does not match the indicator object expression rule, a matching continuation result is generated; otherwise, a non-matching continuation result is generated.

[0096] When the matching result is a match, the subsequent text statement does not introduce new indicator objects, and conditional content consistent with the type of content to be supplemented is identified from the subsequent text statement, the early warning content risk quality control device determines that there is a semantic connection between the indicator objects corresponding to the subsequent text statement and the text condition group identifier, and merges the conditional content consistent with the type of content to be supplemented from the subsequent text statement into the text condition group corresponding to the text condition group identifier. When the matching result is a non-match, or the subsequent text statement introduces new indicator objects, or no conditional content consistent with the type of content to be supplemented is identified from the subsequent text statement, or the text condition group corresponding to the text condition group identifier has recorded a time limit and the subsequent text statement introduces a time limit inconsistent with the recorded time limit, the early warning content risk quality control device determines that there is no semantic connection between the indicator objects corresponding to the subsequent text statement and the text condition group identifier, and establishes a new text condition group based on the conditional content in the subsequent text statement.

[0097] After the risk control device for early warning content completes the grouping, it assigns a text condition group identifier to each text condition group and generates text condition group data. The text condition group data records each text condition group, its identifier, the indicator objects within each text condition group, and one or more of the following: time limit, numerical judgment content, and occurrence status. The text condition group data is used for comparison with the level judgment condition data in S4.

[0098] S4: Based on the level determination condition data, identify the text condition groups in the text condition group data that support the warning level, and obtain the level support data.

[0099] In practice, the early warning content risk control device reads the level determination condition data and the text condition group data, and compares each text condition group with each determination condition in the level determination condition data. The level determination condition data provides the determination conditions corresponding to the current early warning type and level, while the text condition group data provides the condition content in the early warning text that is grouped according to the same condition. The early warning content risk control device uses a single text condition group as the comparison unit and does not merge the condition content from different text condition groups for level support identification.

[0100] The risk control device for early warning content compares each item in the text condition group and the judgment condition according to pre-configured condition comparison rules. These rules include indicator object comparison rules, time limit comparison rules, numerical judgment content comparison rules, and occurrence status comparison rules. The indicator object comparison rules are configured based on the standardized terminology used to represent meteorological indicators in early warning issuance specifications, business rule documents, or local applicable rules, as well as pre-organized synonyms around these standardized terms. This is used to determine whether the indicator object in the text condition group and the indicator object in the judgment condition refer to the same meteorological indicator. The time limit comparison rules are pre-configured based on the textual expressions used to represent statistical periods, forecast periods, or duration periods in early warning issuance specifications, business rule documents, or local applicable rules. This is used to determine whether the time limit in the text condition group and the time limit in the judgment condition represent the same statistical period, forecast period, or duration period. The occurrence status comparison rules are pre-configured based on the status expressions used to represent already occurred, expected to occur, or continue to exist in early warning issuance specifications, business rule documents, or local applicable rules. This is used to determine whether the occurrence status in the text condition group and the occurrence status in the judgment condition represent the same occurrence status.

[0101] The numerical judgment content comparison rules are pre-configured based on the judgment content requirements for corresponding indicator values ​​at different warning levels in the early warning release specifications, business rule documents, or local applicable rules. These rules determine whether the numerical judgment content in the text condition group and the numerical judgment content in the judgment conditions can correspond according to the pre-configured numerical judgment correspondence rules. The early warning content risk quality control device extracts the numerical limitation direction, limitation value, and unit of measurement from both the numerical judgment content in the text condition group and the numerical judgment content in the judgment conditions, respectively, as the processing objects for numerical judgment content comparison. When the unit of measurement used for the limitation value in the text condition group is inconsistent with the unit of measurement used for the limitation value in the judgment conditions, the early warning content risk quality control device unifies the units according to the pre-configured unit conversion rules. The unit conversion rules are stored in the form of a unit conversion table, which includes the indicator object, source unit of measurement, target unit of measurement, conversion coefficient, and conversion offset. The risk control device for early warning content uses the unit of measurement in the indicator object, the record of the main text condition group, and the record of the judgment condition as search fields to find the corresponding item in the unit conversion table. When the corresponding item is found, the limit value under the source unit of measurement is multiplied by the conversion factor and then the conversion offset is added to obtain the limit value under the target unit of measurement. The converted limit value is then used for comparison of the numerical judgment content. When no corresponding item is found, the unit is not uniformly marked and the numerical judgment content is marked as not corresponding in the condition comparison record.

[0102] The preset numerical judgment correspondence rules are pre-configured based on the coverage relationships between "greater than" and "greater than or equal to", "less than" and "less than or equal to", and the correspondence between the limited values ​​in the text condition group and the limited values ​​in the judgment condition under the numerical limitation direction. The numerical limitation directions include greater than, less than, greater than or equal to, and less than or equal to. The preset numerical judgment correspondence rules are as follows: When the numerical limitation direction in the judgment condition is "greater than", if the numerical limitation direction in the text condition group is "greater than", and the limited value in the text condition group is greater than or equal to the limited value in the judgment condition, then the numerical judgment content is determined to correspond; or, if the numerical limitation direction in the text condition group is "greater than or equal to", and the limited value in the text condition group is greater than the limited value in the judgment condition, then the numerical judgment content is determined to correspond. When the numerical limitation direction in the judgment condition is "greater than or equal to", if the numerical limitation direction in the text condition group is "greater than or greater than or equal to", and the limited value in the text condition group is greater than or equal to the limited value in the judgment condition, then the numerical judgment content is determined to correspond. When the numerical constraint in the judgment condition is less than, if the numerical constraint in the main text condition group is less than, and the constraint value in the main text condition group is less than or equal to the constraint value in the judgment condition, then the numerical judgment content is determined to be corresponding; or, if the numerical constraint in the main text condition group is less than or equal to, and the constraint value in the main text condition group is less than the constraint value in the judgment condition, then the numerical judgment content is determined to be corresponding. When the numerical constraint in the judgment condition is less than or equal to, if the numerical constraint in the main text condition group is less than or less than or equal to, and the constraint value in the main text condition group is less than or equal to the constraint value in the judgment condition, then the numerical judgment content is determined to be corresponding.

[0103] The risk control device for early warning content uses the actual recorded condition content types in the judgment conditions as the condition content types for comparison. For condition content types marked with an unrecorded flag in the judgment condition record, the risk control device does not perform a correspondence judgment on the condition content types marked with an unrecorded flag, and writes the condition content types marked with an unrecorded flag into the condition content types not included in the comparison. The condition content types recorded in the condition content types not included in the comparison are neither written into the corresponding condition content types nor into the uncorresponding condition content types.

[0104] The risk control device for early warning content generates a condition comparison record for each text condition group and each judgment condition. The condition comparison record includes the text condition group identifier, judgment condition identifier, indicator object corresponding mark, time limit corresponding mark, numerical judgment content corresponding mark, occurrence status corresponding mark, corresponding condition content type, non-corresponding condition content type, and non-comparison condition content type. The comparison result field is written to the condition comparison record after the comparison result is generated.

[0105] For the condition content types participating in the comparison, if the condition content in the main text condition group corresponds to the condition content in the judgment condition, the early warning content risk quality control device writes a corresponding mark in the condition comparison record and writes the corresponding condition content type into the corresponding condition content type. For condition content types that are actually recorded in the judgment condition but not collected in the main text condition group, or for condition content types that do not correspond to the condition content in the judgment condition, the early warning content risk quality control device writes a non-corresponding mark in the condition comparison record and writes the non-corresponding condition content type into the non-corresponding condition content type.

[0106] The early warning content risk control device generates a comparison result between the text condition group and the judgment condition based on the condition comparison record. When generating the comparison result, the device only reads the condition content types that have corresponding or uncorresponding markers in the condition comparison record, and does not read the condition content types recorded for those not included in the comparison. If all participating condition content types in the condition comparison record have corresponding markers, the device records the comparison result as fully corresponding. If at least one participating condition content type in the condition comparison record has a corresponding marker, and at least one participating condition content type has an uncorresponding marker, the device records the comparison result as partially corresponding. If all participating condition content types in the condition comparison record have uncorresponding markers, the device records the comparison result as uncorresponding. The device then writes the generated comparison result into the corresponding condition comparison record.

[0107] The risk control device for early warning content summarizes the comparison records and results between the same set of conditions in the main text and each judgment condition. If the comparison results between the same set of conditions in the main text and any judgment condition are completely consistent, the risk control device for early warning content determines that the set of conditions in the main text forms level support. If the comparison results between the same set of conditions in the main text and each judgment condition are partially consistent or not consistent, the risk control device for early warning content determines that the set of conditions in the main text does not form level support.

[0108] The risk control device for early warning content generates level support data based on the level support judgment results of each text condition group. For text condition groups that generate level support, the level support data records the text condition group that generates level support, the condition comparison records between the text condition group and the judgment conditions that generate all corresponding comparison results, and one or more judgment conditions that generate all corresponding comparison results with the text condition group. For text condition groups that do not generate level support, the level support data records the text condition group that does not generate level support, the condition comparison records between the text condition group and each judgment condition, the comparison results between the text condition group and each judgment condition, and the content types of corresponding and non-corresponding conditions for each comparison result. This data is used in S5 to supplement the description and identification of text condition groups that do not generate level support.

[0109] S5: Based on the level support data, supplementary descriptions are identified for the text condition groups that do not form level support, and supplementary description data is obtained.

[0110] In practice, the risk control device for early warning content reads from the level support data the text condition groups that do not form level support, the condition comparison records between the text condition groups that do not form level support and each judgment condition, the comparison results, the content types of corresponding conditions, and the content types of non-corresponding conditions. The text condition groups that do not form level support are used to characterize the condition descriptions in the early warning text that fail to form level support within a single text condition group. The condition comparison records are used to characterize the condition correspondence between the text condition groups that do not form level support and the level judgment condition data.

[0111] The risk control device for early warning content identifies text condition groups that lack level support according to pre-configured supplementary description identification rules. These supplementary description identification rules are used to filter out text condition groups that correspond to the judgment condition formation portion but cannot form level support within a single text condition group. The supplementary description identification rules are pre-configured based on the combination of condition content used to organize level judgment conditions in early warning release specifications, business rule documents, or applicable local rules, as well as the text structure in the early warning text used to supplement explanations of condition changes, ongoing status, or incidental effects.

[0112] For text condition groups that do not form level support, the early warning content risk quality control device reads the comparison results between the text condition group and each judgment condition, as well as the condition comparison records that form partially corresponding comparison results. If the comparison result between the text condition group and any judgment condition is partially corresponding, and the type of corresponding condition content in the condition comparison record that forms partially corresponding comparison results includes indicator objects, and the type of corresponding condition content also includes at least one of time limits, numerical judgment content, or occurrence status, while the type of non-corresponding condition content includes at least one of time limits, numerical judgment content, or occurrence status of the actual record of the judgment condition, then the early warning content risk quality control device identifies the text condition group pointed to by the text condition group identifier in the condition comparison record that forms partially corresponding comparison results as a supplementary description text condition group. The supplementary description text condition group is used to characterize text condition groups that partially correspond to judgment conditions, do not form level support within the scope of a single text condition group, but have included some judgment condition content, and are separately prompted as objects in level consistency quality control.

[0113] For text condition groups that do not form level-based support, if the comparison results between the text condition group and each judgment condition are all non-corresponding, the early warning content risk control device will not identify the text condition group that does not form level-based support as a supplementary description text condition group. For text condition groups that do not form level-based support but have partial corresponding comparison results with judgment conditions, the early warning content risk control device will, according to the supplementary description identification rules, judge each condition comparison record that forms a partial corresponding comparison result to see if it conforms to the supplementary description identification rules. If the corresponding condition content type in the condition comparison record that forms a partial corresponding comparison result does not include the indicator object, or the corresponding condition content type only includes the indicator object, or the non-corresponding condition content type does not include any of the time limit, numerical judgment content, or occurrence status of the actual record of the judgment condition, the early warning content risk control device will determine that the condition comparison record that forms a partial corresponding comparison result does not conform to the supplementary description identification rules. If there is no condition comparison record that conforms to the supplementary description identification rules, the early warning content risk control device will not identify the text condition group pointed to by the text condition group identifier in the condition comparison record that forms a partial corresponding comparison result as a supplementary description text condition group.

[0114] The risk control device for early warning content generates supplementary description data based on the identification results of the supplementary description. This supplementary description data records the supplementary description text condition group, one or more judgment conditions corresponding to the part forming the supplementary description text condition group, a condition comparison record between the supplementary description text condition group and the judgment conditions corresponding to the part forming the condition group, the content types of corresponding and non-corresponding conditions in the condition comparison record, and one or more original text locations. This data is used in S6 to combine level support data and level judgment condition data to perform quality control on the consistency between the early warning level and the early warning text.

[0115] S6: Combining the level support data, supplementary description data, and level determination condition data, perform quality control on the level consistency between the warning level and the warning text to obtain the level consistency quality control results.

[0116] In practice, the risk control device for early warning content reads the level support data, supplementary description data, and level determination condition data. The level support data is used to provide the judgment result of whether the text condition group forms level support. The supplementary description data is used to provide the supplementary description text condition group that does not form level support but corresponds to the part that forms the determination condition. The level determination condition data is used to provide the determination conditions corresponding to the current early warning type and the current early warning level.

[0117] The risk control device for early warning content reads the textual condition groups that form level support from the level support data. If the level support data records textual condition groups that form level support, the device generates a level basis formation marker and records the textual condition groups that form level support, as well as the judgment conditions that form all corresponding comparison results with the textual condition groups. If the level support data does not record textual condition groups that form level support, the device generates a level basis insufficiency marker and records the judgment conditions in the level judgment condition data that do not form all corresponding comparison results with any textual condition group as level basis insufficiency judgment conditions. Level basis insufficiency judgment conditions are used to characterize judgment conditions that do not form all corresponding comparison results with any textual condition group in the early warning text.

[0118] The risk control device for early warning content determines whether a supplementary description condition group exists based on the supplementary description data. If the supplementary description data records a supplementary description condition group, the device generates a supplementary description prompt marker and records the supplementary description condition group, the judgment conditions corresponding to the part forming the supplementary description condition group, the condition comparison record between the supplementary description condition group and the judgment conditions corresponding to the part forming the supplementary description condition group, and the types of content with corresponding and non-corresponding conditions in the condition comparison record. The device further determines whether a supplementary description condition group exists based on the supplementary description data. If at least one supplementary description condition group is recorded in the supplementary description data, the supplementary description status is recorded as "supplementary description exists," and a supplementary description prompt marker is generated; if no supplementary description condition group is recorded in the supplementary description data, the supplementary description status is recorded as "no supplementary description." The supplementary description status is written into the level consistency quality control result.

[0119] The early warning content risk quality control device generates a level consistency quality control result based on level basis formation markers, level basis insufficiency markers, supplementary description prompt markers, and no supplementary description prompt markers. The device determines the level basis status value based on either the level basis formation marker or the level basis insufficiency marker. When a level basis formation marker is generated, the device records the level basis status value as "level basis formed." When a level basis insufficiency marker is generated, the device records the level basis status value as "level basis insufficiency." When the level basis status value is recorded as "level basis formed," the device writes the text condition group that forms the level support into the level consistency quality control result. When the level basis status value is recorded as "level basis insufficiency," the device writes the level basis insufficiency judgment condition into the level consistency quality control result.

[0120] When a supplementary description prompt is generated, the risk control device for early warning content will write the supplementary description text condition group, the judgment condition corresponding to the part that forms the supplementary description text condition group, the condition comparison record between the supplementary description text condition group and the judgment condition corresponding to the part that forms the supplementary description text condition group, the content type of the corresponding condition in the condition comparison record, the content type of the non-corresponding condition, and one or more original text positions into the supplementary description text condition group field, the judgment condition field, the condition comparison record field, the corresponding type field, the non-corresponding type field, and the original text position field in the quality control prompt content, respectively, to indicate that the supplementary description text condition group has not formed a level support; when the supplementary description status is no supplementary description, the risk control device for early warning content will not write into the quality control prompt content.

[0121] If a level basis formation marker is generated and a marker indicating no supplementary description is generated, the level consistency quality control result includes the level basis status with the value of the level basis formation. If a level basis formation marker is generated and a marker indicating supplementary description is generated, the level consistency quality control result includes the level basis status with the value of the level basis formation, and also includes quality control prompt content. If a level basis insufficiency marker is generated and a marker indicating no supplementary description is generated, the level consistency quality control result includes the level basis status with the value of the level basis insufficiency. If a level basis insufficiency marker is generated and a marker indicating supplementary description is generated, the level consistency quality control result includes the level basis status with the value of the level basis insufficiency, and also includes quality control prompt content.

[0122] The level consistency quality control result includes one or more of the following: level basis status, supplementary description status, text condition group forming level support, level basis insufficient judgment condition, supplementary description prompt mark, supplementary description text condition group, and quality control prompt content. The level consistency quality control result is used to characterize the level consistency between the warning level in the warning data to be quality controlled and the warning text. The warning content risk quality control device returns the level consistency quality control result to the warning content publishing system, which then displays the level basis status and quality control prompt content based on the level consistency quality control result.

[0123] In one possible implementation, when executing S2, the risk control device for early warning content performs standardized statement location, condition content extraction, connection expression identification, and judgment condition record generation on the level judgment statements in the early warning level specification data to form level judgment condition data. This level judgment condition data is used to convert the judgment requirements for the corresponding early warning level in the early warning release specification into structured condition data that can participate in the comparison processing in S4.

[0124] In practice, the risk control device for early warning content reads the early warning type and level from the early warning data to be controlled, and uses the early warning type and level as search criteria to find level determination statements in the early warning level specification data that are consistent with the early warning type and level. When the early warning level specification data is stored in the form of specification clauses, specification paragraphs, or specification statements, the risk control device for early warning content reads the specification clause number, specification paragraph number, or specification statement number associated with the same early warning type and the same early warning level, and uses the text content pointed to by the specification clause number, specification paragraph number, or specification statement number as the level determination statement.

[0125] The risk control device for early warning content performs text standardization processing on the level determination statements. This standardization includes unifying meteorological indicator names to those in the standardized condition dictionary, unifying units of measurement to the standard units of measurement recorded in early warning issuance specifications, business rule documents, or applicable local rules, and unifying numerical direction constraints to greater than, greater than or equal to, or less than or less than or equal to, while retaining the original expression of the constraint values. The risk control device also retains the source text location in the level determination statements. The standardized text is used for condition content extraction, and the source text location is used to locate the source position of the determination conditions in the early warning level standard data.

[0126] The risk control device for early warning content extracts the indicator object, time limit, numerical judgment content, and occurrence status from the level determination statement based on a pre-configured standard condition dictionary. The standard condition dictionary is pre-configured according to early warning issuance specifications, business rule documents, or applicable local rules. The indicator object corresponds to the name of the meteorological indicator or its abbreviation; the time limit corresponds to the statistical period, forecast period, or duration; the numerical judgment content corresponds to the direction, value, and unit of measurement of the numerical limit; and the occurrence status corresponds to the requirement that it has occurred, is expected to occur, or is ongoing.

[0127] The connection expressions in the risk control device's judgment statement for warning content identification level. Connection expressions include joint validity expressions and alternative validity expressions. Joint validity expressions indicate that multiple conditions jointly define the same judgment condition, while alternative validity expressions indicate that any one of the multiple segments can independently represent the judgment requirement for the corresponding warning level to be established. Joint validity expressions and alternative validity expressions are pre-configured based on the conditional connection terms in the warning release specifications, business rule documents, or applicable local rules.

[0128] When multiple conditions are connected by a common conditional expression, the early warning content risk control device consolidates these conditions into a single judgment condition. When the level judgment statement contains an alternative conditional expression, the early warning content risk control device uses this alternative expression as the dividing boundary to divide the level judgment statement into multiple segments, and reads the conditional content from each segment. If a segment includes an indicator object, and also includes one or more of the following: a time limit, numerical judgment content, or occurrence state that jointly defines the judgment requirement for the corresponding early warning level, the early warning content risk control device consolidates the segment containing the indicator object and one or more of the following: a time limit, numerical judgment content, or occurrence state that jointly defines the judgment requirement for the corresponding early warning level, into a single judgment condition.

[0129] The risk control device for early warning content assigns a judgment condition identifier to each judgment condition and generates a judgment condition record. The judgment condition record includes the judgment condition identifier, indicator object, time limit, numerical judgment content, occurrence status, and source text location. If the time limit, numerical judgment content, or occurrence status is not actually recorded in the judgment condition, the risk control device for early warning content writes a "not recorded" mark in the judgment condition record. The "not recorded" mark is used in S4 to exclude condition content types that did not actually participate in the judgment requirements for the corresponding early warning level, avoiding the treatment of condition content types not required by the specification as non-corresponding content.

[0130] The risk control device for early warning content aggregates the judgment condition records under the same early warning type and level, forming level judgment condition data. This data includes the early warning type, early warning level, set of judgment condition records, and source information for the judgment conditions. The set of judgment condition records is used for comparison with the data in the main text condition group in S4. The source information for the judgment conditions is used to locate the source of the insufficient judgment conditions in the early warning level specification data when there are insufficient judgment conditions based on the recorded level.

[0131] In one possible implementation, when executing S3, the risk control device for early warning content performs sentence segmentation, condition content identification, grouping of identical conditions, and generation of text condition group records on the early warning text to obtain text condition group data. This text condition group data is used to convert the natural language condition descriptions in the early warning text into structured text condition data that can participate in the comparison processing in S4.

[0132] In practice, the risk control device for early warning content reads the main text of the early warning data to be controlled and segments it according to pre-defined sentence segmentation boundaries. These boundaries include periods, semicolons, line breaks, and segmentation markers in the early warning release template. The device assigns sentence identifiers to the segmented sentences and records their start and end positions within the early warning text.

[0133] The early warning content risk control device performs text recognition processing on the main text statements according to the conditional content recognition rules. These rules include rules for expressing indicator objects, time limits, numerical judgments, and occurrence states. The device then organizes the identified conditional content into a main text conditional content record. This record includes the content type, original text content, normalized text, original text location, and the corresponding statement identifier. The normalized text is used for conditional comparison in S4, while the original text content and location are used for generating quality control prompts in S6.

[0134] The risk control device for early warning content uses indicator objects as the core grouping mechanism. It reads the time limits, numerical judgments, and occurrence states of content that are in the same statement as the indicator object or have a semantic connection, and processes the read conditional content as the grouping object for the main text condition group. The semantic connection relationship is determined based on semantic connection expressions, which include textual expressions indicating content continuation, partial supplementation, or continuous explanation. The semantic connection expressions are pre-configured according to the early warning release specifications, business rule documents, or commonly used early warning text description methods in applicable local rules.

[0135] When one or more conditions—time limits, numerical judgments, or occurrence states—in the same statement jointly define the same indicator object, and do not form another set of independent conditions, the early warning content risk control device will group the indicator object and the conditions used to define the same indicator object into the same text condition group. When the same statement forms two or more sets of independent conditions around the same indicator object, and each set of conditions constitutes an independent condition around different time limits, different numerical judgments, or different occurrence states, the early warning content risk control device will establish separate text condition groups.

[0136] For adjacent text statements, the early warning content risk control device reads the text condition group already established in the previous text statement and checks whether the text condition group is missing at least one of the following conditions: time limit, numerical judgment content, or occurrence status. If the text condition group in the previous text statement is missing at least one of the following conditions: time limit, numerical judgment content, or occurrence status, and the subsequent text statement does not introduce a new indicator object, and the condition content in the subsequent text statement can supplement the missing condition content type in the text condition group of the previous text statement, the early warning content risk control device will merge the condition content in the subsequent text statement into the text condition group of the previous text statement.

[0137] If a time limit has been recorded in the condition group of a preceding statement, and a subsequent statement introduces a time limit that is inconsistent with the recorded time limit, the risk control device for early warning content will not incorporate the condition content of the subsequent statement into the condition group of the preceding statement. If a subsequent statement introduces a new indicator object, or if the condition content of the subsequent statement cannot supplement the missing condition content type in the condition group of the preceding statement, the risk control device for early warning content will create a new condition group.

[0138] The risk control device for early warning content assigns a text condition group identifier to each text condition group and generates a text condition group record. The text condition group record includes the text condition group identifier, indicator object, time limit, numerical judgment content, occurrence status, source statement identifier, and original text location. If a time limit, numerical judgment content, or occurrence status is not included in the text condition group, the risk control device can write a non-collection mark to the text condition group record. This non-collection mark is used in S4 to assist in identifying the type of condition content that is actually recorded but not included in the text condition group. If no non-collection mark is written, the risk control device identifies the type of condition content that is actually recorded but not included in the text condition group based on whether the time limit, numerical judgment content, or occurrence status is not recorded in the text condition group record.

[0139] The risk control device for early warning content summarizes the records of the main text condition group, forming the main text condition group data. The main text condition group data includes a set of main text condition group records and source information for the main text condition content. The set of main text condition group records is used for condition comparison with the level determination condition data in S4, and the source information for the main text condition content is used to generate quality control prompts in S6.

[0140] In one possible implementation, when executing S4, the risk control device for early warning content generates a condition comparison record based on the level determination condition data and the text condition group data, and then generates level support data based on the condition comparison record. The level support data is used to distinguish between text condition groups in the early warning text that form level support, and text condition groups that do not form level support but still need to participate in the supplementary description identification in S5.

[0141] In practice, the early warning content risk control device reads the set of judgment condition records from the level judgment condition data and the set of text condition group records from the text condition group data. The device compares a single text condition group record and a single judgment condition record as a set of comparison objects, and reads the text condition group identifier and judgment condition identifier to generate condition comparison records. The early warning content risk control device does not merge the condition content from different text condition group records and compare it with the same judgment condition record.

[0142] The risk control device for early warning content reads the unrecorded markers in the judgment condition record and excludes the condition content types identified by the unrecorded markers from the comparison. The condition content types included in the comparison are the indicator objects actually recorded in the judgment condition record, as well as one or more of the time limits, numerical judgment content, or occurrence states actually recorded in the judgment condition record. For condition content types without recorded markers, the risk control device writes the unrecorded markers into the non-compared condition content types and does not write the condition content types recorded in the non-compared condition content types into the already corresponding or uncorresponding condition content types.

[0143] The risk control device for early warning content performs a step-by-step comparison of the conditions and content types involved in the comparison according to the condition comparison rules. The indicator object comparison rule is used to determine whether the indicator object in the main condition group record and the indicator object in the judgment condition record point to the same meteorological indicator; the time limit comparison rule is used to determine whether the time limit in the main condition group record and the time limit in the judgment condition record represent the same statistical period, forecast period, or duration; the numerical judgment content comparison rule is used to determine whether the numerical judgment content in the main condition group record and the numerical judgment content in the judgment condition record correspond to each other; and the occurrence status comparison rule is used to determine whether the occurrence status in the main condition group record and the occurrence status in the judgment condition record represent the same status that has occurred, is expected to occur, or is ongoing.

[0144] The numerical judgment comparison is performed according to preset numerical judgment correspondence rules. The early warning content risk quality control device extracts the numerical limitation direction, limitation value, and unit of measurement from the numerical judgment content in the text condition group record and the judgment condition record, respectively. When the unit of measurement used for the limitation value in the text condition group record is inconsistent with the unit of measurement used for the limitation value in the judgment condition record, the early warning content risk quality control device completes the unit unification according to the unit conversion rules. The unit conversion rules are pre-configured based on the correspondence of the units of measurement used for the same meteorological indicator in the early warning release specifications, business rule documents, or local applicable rules. After unit unification, the early warning content risk quality control device determines whether the numerical judgment content in the text condition group record corresponds to the numerical judgment content in the judgment condition record according to the limitation coverage relationship between greater than, greater than or equal to, less than, and less than or equal to.

[0145] The risk control device for early warning content generates a condition comparison record. This record includes the text condition group identifier, judgment condition identifier, indicator object correspondence marker, time limit correspondence marker, numerical judgment content correspondence marker, occurrence status correspondence marker, corresponding condition content type, non-corresponding condition content type, and non-comparison condition content type. For condition content types that form a corresponding condition, the risk control device writes the corresponding marker in the condition comparison record and adds the corresponding condition content type to the "corresponding condition content type" field. For condition content types that are actually recorded in the judgment condition record but not included in the text condition group record, or for condition content in the text condition group record that does not form a corresponding condition content type with the condition content in the judgment condition record, the risk control device writes a "non-corresponding" marker in the condition comparison record and adds the non-corresponding condition content type to the "non-corresponding condition content type" field.

[0146] The early warning content risk control device generates comparison results based on the condition comparison records. The device counts the number of condition content types participating in the comparison in the condition comparison records, and also counts the number of condition content types marked with corresponding tags, excluding condition content types recorded for those not participating in the comparison. When the number of condition content types marked with corresponding tags matches the number of condition content types participating in the comparison, the device records the comparison result as a complete match; when the number of condition content types marked with corresponding tags is greater than zero and less than the number of condition content types participating in the comparison, the device records the comparison result as a partial match; when the number of condition content types marked with corresponding tags is zero, the device records the comparison result as no match. The early warning content risk control device then writes the generated comparison results into the corresponding condition comparison records.

[0147] The risk control device for early warning content uses the text condition group identifier as the summary field to summarize the condition comparison records between the same text condition group record and each judgment condition record. If the comparison results between the text condition group record and any judgment condition record are all correct, the risk control device for early warning content determines the text condition group record as a text condition group that forms level support. If the comparison results between the text condition group record and each judgment condition record are all partially correct or not correct, the risk control device for early warning content determines the text condition group record as a text condition group that does not form level support.

[0148] The risk control device for early warning content generates level support data based on the level support judgment results. Level support data includes the text condition groups that generate level support, condition comparison records between the text condition groups that generate level support and the judgment conditions that generate all corresponding comparison results, text condition groups that do not generate level support, condition comparison records between the text condition groups that do not generate level support and each judgment condition, comparison results, content types of corresponding conditions, and content types of non-corresponding conditions. Level support data is used in S5 to identify and supplement text condition groups for description, and in S6 to generate level basis formation markers or level basis insufficiency markers.

[0149] In one possible implementation, when the early warning content risk quality control device executes S5, it identifies supplementary description text condition groups based on the text condition groups that do not form level support and condition comparison records in the level support data, and forms supplementary description data. The supplementary description data is used to use the condition descriptions in the early warning text that correspond to the judgment condition forming part but do not form level support as the quality control prompt objects in S6.

[0150] In practice, the early warning content risk control device reads the text condition groups that do not form level support from the level support data, and reads the condition comparison records between the text condition groups that do not form level support and each judgment condition. The early warning content risk control device uses the text condition group identifier as a filtering field to filter the condition comparison records from the condition comparison records, and uses the filtered condition comparison records to determine whether the text condition groups that do not form level support belong to the supplementary description text condition groups.

[0151] The risk control device for early warning content judges each selected condition comparison record according to the supplementary description identification rules. The supplementary description identification rules include requirements corresponding to the indicator object, requirements where the limiting conditions are already corresponding, and requirements where the limiting conditions are not corresponding. The content type of the requirement where the indicator object is already corresponding includes the indicator object; the content type of the requirement where the limiting conditions are already corresponding also includes at least one of time limit, numerical judgment content, or occurrence status; the content type of the requirement where the limiting conditions are not corresponding includes at least one of the actual recorded time limit, numerical judgment content, or occurrence status of the judgment condition.

[0152] When a record comparing the same condition simultaneously meets the requirements of corresponding indicator objects, corresponding limiting conditions, and non-corresponding limiting conditions, the risk control device for early warning content determines that the record comparing the condition meets the supplementary description identification rules, and identifies the text condition group pointed to by the text condition group identifier in the record comparing the condition as the supplementary description text condition group. The supplementary description text condition group is used to characterize the condition descriptions in the early warning text that have partially corresponded to the judgment conditions, but have not formed level support within a single text condition group.

[0153] When the comparison results between a text condition group that lacks level-based support and each judgment condition are all non-corresponding, the early warning content risk control device will not identify the text condition group that lacks level-based support as a supplementary description text condition group. When a text condition group that lacks level-based support has partially corresponding condition comparison records, but none of the partially corresponding condition comparison records meet the supplementary description identification rules, the early warning content risk control device will not identify the text condition group that lacks level-based support as a supplementary description text condition group.

[0154] The risk control device for early warning content writes the condition comparison records that meet the supplementary description identification rules into the supplementary description data. The supplementary description data includes the supplementary description text condition group, the judgment conditions corresponding to the forming part of the supplementary description text condition group, the condition comparison records between the supplementary description text condition group and the judgment conditions corresponding to the forming part, the content types of the corresponding conditions, the content types of the uncorresponding conditions, and one or more original text positions. One or more original text positions are written according to the original text positions in the text condition group data, and are used to form the quality control prompt content in S6.

[0155] Supplementary description data is used to generate a supplementary description prompt mark in S6. When the supplementary description data records at least one supplementary description text condition group, the early warning content risk quality control device generates a supplementary description prompt mark in S6 and writes the prompt content indicating that there is no level support between the supplementary description text condition group and the corresponding judgment condition of the forming part into the quality control prompt content. When the supplementary description data does not record a supplementary description text condition group, the early warning content risk quality control device generates a no-supplementary-description prompt mark in S6 and does not write it into the quality control prompt content.

[0156] See Figure 3 As shown in the figure, this application provides a risk control system for early warning content based on multi-granularity semantic analysis. The risk control system for early warning content based on multi-granularity semantic analysis includes a data acquisition module, a level judgment condition formation module, a text condition group formation module, a level support data formation module, a supplementary description data formation module, and a quality control result generation module.

[0157] The data acquisition module is used to acquire the quality control early warning data and the early warning level specification data. The quality control early warning data includes the early warning type, early warning level, and early warning text. The early warning level specification data is managed according to the early warning type and early warning level, and records the level judgment statements used to characterize the judgment requirements for the corresponding early warning level under different early warning types and levels. The data acquisition module transmits the early warning type and early warning level to the level judgment condition formation module, transmits the early warning text to the text condition group formation module, and transmits the early warning level specification data to the level judgment condition formation module.

[0158] The level determination condition formation module combines the warning type and warning level to extract the corresponding determination conditions from the warning level specification data, thus obtaining level determination condition data. In practice, the module uses the warning type and warning level as search criteria to find level determination statements in the warning level specification data that match the warning type and warning level. It then performs text normalization on the level determination statements, extracting the indicator object, time limit, numerical determination content, and occurrence status. It identifies common-validity expressions and alternative-validity expressions in the level determination statements. Based on common-validity expressions, it organizes multiple conditions that jointly define the same determination condition into a single determination condition. Based on alternative-validity expressions, it organizes segments that can individually represent the determination requirements for the corresponding warning level into determination conditions. The module assigns a determination condition identifier to each determination condition and forms a determination condition record including the determination condition identifier, indicator object, time limit, numerical determination content, occurrence status, and source text location. If the time limit, numerical determination content, or occurrence status is not actually recorded in the determination condition, the module writes a "not recorded" mark in the determination condition record. The level determination condition formation module summarizes the determination condition records under the same warning type and the same warning level, forms level determination condition data, and transmits the level determination condition data to the level support data formation module and the quality control result generation module.

[0159] The text condition group formation module is used to identify the conditional content in the warning text and group the conditional content representing the same condition to obtain text condition group data. In specific implementation, the text condition group formation module segments the warning text according to pre-defined sentence segmentation boundaries; according to conditional content identification rules, it identifies the indicator object, time limit, numerical judgment content, and occurrence status from the text sentences; and organizes the identified conditional content into text conditional content records. The text condition group formation module uses the indicator object as the core of grouping, reading the time limit, numerical judgment content, and occurrence status that are in the same sentence or have a semantic connection with the indicator object, and using the read conditional content as the processing object for text condition grouping. When one or more of the time limit, numerical judgment content, or occurrence status conditions in the same sentence jointly define the same indicator object, and do not form another independent set of conditional content, the text condition group formation module groups the indicator object and the conditional content defining the same indicator object into the same text condition group; when the same sentence forms two or more independent sets of conditional content around the same indicator object, the text condition group formation module establishes separate text condition groups. The text condition group formation module assigns a text condition group identifier to each text condition group and generates a text condition group record that includes the text condition group identifier, indicator object, time limit, numerical judgment content, occurrence status, source statement identifier, and original text location. The text condition group formation module summarizes the text condition group records to form text condition group data and transmits the text condition group data to the level support data formation module.

[0160] The level support data formation module is used to identify the text condition groups in the text condition group data that support the warning level, based on the level judgment condition data and the text condition group data, to obtain the level support data. In specific implementation, the level support data formation module reads the judgment condition record set from the level judgment condition data and the text condition group record set from the text condition group data; it uses a single text condition group record and a single judgment condition record as a comparison object, reading the text condition group identifier and the judgment condition identifier to generate condition comparison records. The level support data formation module reads the unrecorded markers in the judgment condition records and excludes the condition content types identified by the unrecorded markers from participating in the comparison. The level support data formation module writes the condition content types identified by the unrecorded markers into the uncompared condition content types; the condition content types recorded in the uncompared condition content types are not written into the corresponding condition content types, nor into the uncorresponding condition content types. The level-supported data generation module performs a step-by-step comparison of the participating indicator objects, time limits, numerical judgment content, and occurrence status according to the condition comparison rules. For corresponding condition content types, a corresponding mark is written into the condition comparison record, and the corresponding condition content type is written into the corresponding condition content type. For condition content types that are actually recorded in the judgment condition record but not included in the main text condition group record, or for condition content in the main text condition group record that does not form a corresponding condition content type with the condition content in the judgment condition record, a non-corresponding mark is written into the condition comparison record, and the non-corresponding condition content type is written into the non-corresponding condition content type. The condition comparison record includes the main text condition group identifier, judgment condition identifier, indicator object corresponding mark, time limit corresponding mark, numerical judgment content corresponding mark, occurrence status corresponding mark, corresponding condition content type, non-corresponding condition content type, and non-participating condition content type. The level-supported data generation module generates comparison results that are all corresponding, partially corresponding, or non-corresponding based on the condition comparison record and writes the comparison results into the corresponding condition comparison record. When the comparison result between the main text condition group record and any judgment condition record is a complete match, the level support data formation module determines the main text condition group record as a main text condition group that forms level support. When the comparison result between the main text condition group record and each judgment condition record is a partial match or no match, the level support data formation module determines the main text condition group record as a main text condition group that does not form level support. The level support data formation module generates level support data based on the level support judgment result and transmits the level support data to the supplementary description data formation module and the quality control result generation module.

[0161] The supplementary description data formation module is used to identify supplementary descriptions for text condition groups that lack level support based on level support data, thereby obtaining supplementary description data. Specifically, the module reads text condition groups that lack level support from the level support data and reads the condition comparison records between these groups and each judgment condition. Using the text condition group identifier as a filtering field, the module filters condition comparison records where the comparison results partially correspond, and uses these filtered records to determine whether a text condition group lacking level support belongs to a supplementary description text condition group. The module then judges each filtered condition comparison record according to the supplementary description identification rules. These rules include requirements for corresponding indicator objects, requirements for corresponding limiting conditions, and requirements for non-corresponding limiting conditions. For requirements for corresponding indicator objects, the content type of the corresponding limiting condition includes the indicator object. For requirements for corresponding limiting conditions, the content type also includes at least one of time limits, numerical judgment content, or occurrence status. For requirements for non-corresponding limiting conditions, the content type includes at least one of the actual record of the judgment condition, such as time limits, numerical judgment content, or occurrence status. When a record under the same condition meets the requirements of corresponding indicator objects, corresponding limiting conditions, and non-corresponding limiting conditions, the supplementary description data formation module determines that the record under the condition meets the supplementary description identification rules and identifies the text condition group pointed to by the text condition group identifier in the record under the condition as the supplementary description text condition group.

[0162] When the comparison results between a text condition group without level support and each judgment condition are all non-corresponding, the supplementary description data formation module will not identify the text condition group without level support as a supplementary description text condition group. When a text condition group without level support has partially corresponding condition comparison records, but none of the partially corresponding condition comparison records meet the supplementary description identification rules, the supplementary description data formation module will not identify the text condition group without level support as a supplementary description text condition group. The supplementary description data formation module writes the condition comparison records that meet the supplementary description identification rules into the supplementary description data. The supplementary description data includes the supplementary description text condition group, the judgment conditions corresponding to the forming part of the supplementary description text condition group, the condition comparison records between the supplementary description text condition group and the judgment conditions corresponding to the forming part, the content types of the corresponding conditions, the content types of the non-corresponding conditions, and one or more original text positions. The supplementary description data formation module transmits the supplementary description data to the quality control result generation module.

[0163] The quality control result generation module combines level support data, supplementary descriptive data, and level determination condition data to perform quality control on the consistency between the warning level and the warning text, obtaining a level consistency quality control result. In specific implementation, the quality control result generation module reads the text condition groups that form level support from the level support data. When the level support data records text condition groups that form level support, the quality control result generation module generates a level basis formation marker and records the text condition groups that form level support, as well as the determination conditions that form all corresponding comparison results with the text condition groups. When the level support data does not record text condition groups that form level support, the quality control result generation module generates a level basis insufficiency marker and records the determination conditions in the level determination condition data that do not form all corresponding comparison results with any text condition group as level basis insufficiency determination conditions.

[0164] The quality control result generation module determines whether a supplementary description text condition group exists based on the supplementary description data. If the supplementary description data records a supplementary description text condition group, the module generates a supplementary description prompt marker and records the supplementary description text condition group, the judgment conditions corresponding to the part forming the supplementary description text condition group, the condition comparison record between the supplementary description text condition group and the judgment conditions corresponding to the part forming the supplementary description text condition group, and the content types of corresponding and non-corresponding conditions in the condition comparison record. If the supplementary description data does not record a supplementary description text condition group, the module generates a no-supplementary-description prompt marker.

[0165] The quality control result generation module determines the level basis status value based on whether a level basis formation mark or a level basis insufficiency mark is generated. When a level basis formation mark is generated, the quality control result generation module records the level basis status value as "level basis formed" and writes the text condition group that forms level support into the level consistency quality control result. When a level basis insufficiency mark is generated, the quality control result generation module records the level basis status value as "level basis insufficiency" and writes the level basis insufficiency judgment condition into the level consistency quality control result. When a supplementary description prompt mark is generated, the quality control result generation module writes the supplementary description text condition group, the judgment condition corresponding to the forming part of the supplementary description text condition group, the condition comparison record between the supplementary description text condition group and the judgment condition corresponding to the forming part, the corresponding condition content type in the condition comparison record, the uncorresponding condition content type, and one or more original text positions into the quality control prompt content to indicate that the supplementary description text condition group has not formed level support. When no supplementary description prompt mark is generated, the quality control result generation module does not write the quality control prompt content. The quality control result generation module writes one or more of the following into the level consistency quality control result: level-based status, level-supporting text condition groups, level-based insufficient judgment conditions, supplementary description prompt markers, supplementary description text condition groups, and quality control prompt content.

[0166] In one possible implementation, the data acquisition module, level determination condition formation module, text condition group formation module, level supporting data formation module, supplementary description data formation module, and quality control result generation module are deployed on the same server, the same early warning content risk quality control device, or the same early warning content publishing system. Alternatively, these modules can be deployed on multiple computing devices connected via a communication network. Different deployment methods do not alter the data flow relationships between the early warning data to be quality controlled, the early warning level specification data, the level determination condition data, the text condition group data, the level supporting data, the supplementary description data, and the level consistency quality control results.

[0167] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for risk control of early warning content based on multi-granularity semantic analysis, characterized in that, include: Obtain the quality control early warning data and early warning level specification data, wherein the quality control early warning data includes the early warning type, early warning level, and early warning text; Combining the warning type and the warning level, the judgment conditions corresponding to the warning type and the warning level are extracted from the warning level specification data to obtain the level judgment condition data; The conditional content in the warning text is identified, and the conditional content used to represent the same condition is grouped to obtain the text conditional group data; Based on the level determination condition data, identify the text condition groups in the text condition group data that support the warning level, and obtain the level support data; Based on the level support data, supplementary descriptions are identified for text condition groups that do not form level support, and supplementary description data is obtained. By combining the level support data, the supplementary description data, and the level determination condition data, quality control is performed on the level consistency between the warning level and the warning text to obtain the level consistency quality control result.

2. The method for risk control of early warning content based on multi-granularity semantic analysis according to claim 1, characterized in that, The method for obtaining the text condition group data includes: The warning text is segmented according to the pre-defined sentence segmentation boundaries to obtain the main text sentences; According to the pre-configured conditional content recognition rules, the main text statement is subjected to text recognition to obtain multiple conditional contents, and the content type of each conditional content is recorded. The content type is determined from the indicator object, time limit, numerical judgment content and occurrence status. Filter the content of the multiple conditional contents that is of the type of indicator object, and extract the indicator objects from the filtered conditional contents; Based on the extracted indicator objects, the conditional content that is in the same sentence as the indicator object or has a semantic connection relationship is grouped to form the main text conditional group; Assign a text condition group identifier to the text condition group to form text condition group data.

3. The method for risk control of early warning content based on multi-granularity semantic analysis according to claim 2, characterized in that, The method for determining the existence of semantic succession relationships includes: In adjacent text statements, read the text condition groups that have been established in the text statements that are sorted first, and read the text condition group identifier and index object corresponding to the text condition group. Using the text condition group identifier as the processing unit, check whether the text condition group has recorded the time limit, numerical judgment content and occurrence status, and determine the unrecorded time limit, numerical judgment content or occurrence status as the content type to be supplemented. According to the pre-configured semantic continuation expression, the conditional content in the subsequent text statements is semantically matched with the index object corresponding to the text condition group identifier to obtain the continuation matching result. When the matching result is that the following text statement does not introduce new indicator objects, and condition content consistent with the type of content to be supplemented is identified from the following text statement, it is determined that the following text statement and the indicator object corresponding to the text condition group identifier have a semantic connection relationship.

4. The method for risk control of early warning content based on multi-granularity semantic analysis according to claim 1, characterized in that, The method for obtaining the level support data includes: Extract judgment condition records from the level judgment condition data, and extract text condition group records from the text condition group data; Using a single text condition group record and a single decision condition record as comparison objects, the condition content type of the actual record in the decision condition record is read as the condition content type to be compared. For each condition content type involved in the comparison, the condition content in the main text condition group record and the judgment condition record is compared; for the condition content types that form a corresponding condition, a corresponding mark and a corresponding condition content type are written; for the condition content types that do not form a corresponding condition, a non-corresponding mark and a non-corresponding condition content type are written, and a condition comparison record is generated. Based on the conditions, comparison records are generated to produce comparison results that show all corresponding records, some corresponding records, or no corresponding records. When the comparison result between the main text condition group record and any decision condition record is completely consistent, the main text condition group record is marked as forming level support; when the comparison result between the main text condition group record and each decision condition record is partially consistent or not consistent, the main text condition group record is marked as not forming level support; the marking results, condition comparison records and comparison results are summarized to form level support data.

5. The method for risk control of early warning content based on multi-granularity semantic analysis according to claim 4, characterized in that, The method for obtaining supplementary descriptive data includes: Read the text condition groups that do not form level support, along with the corresponding condition comparison records and comparison results, from the level support data. From the condition comparison records and the comparison results, filter out the condition comparison records that correspond to some of the comparison results to obtain the condition comparison records to be identified; According to the pre-configured supplementary description identification rules, the corresponding and non-corresponding condition content types in the comparison record of the conditions to be identified are judged; When the corresponding condition content type includes an indicator object, and also includes at least one of time limit, numerical judgment content, or occurrence status, and the uncorresponding condition content type includes at least one of time limit, numerical judgment content, or occurrence status of the actual record of the judgment condition, the text condition group forming the comparison record of the condition to be identified will be identified as a supplementary description text condition group. Write the supplementary description text condition group, the judgment condition for forming the comparison record of the condition to be identified, the comparison record of the condition to be identified, the content type of the corresponding condition, and the content type of the uncorresponding condition into the supplementary description data.

6. The method for risk control of early warning content based on multi-granularity semantic analysis according to claim 5, characterized in that, The method for obtaining the level consistency quality control results includes: Read the tagging results, condition comparison records, and comparison results from the level support data, and read the supplementary description text condition groups from the supplementary description data; Filter the text condition groups marked as forming level support from the marked results. If the filtering results are not empty, generate level basis forming tags and write the filtered text condition groups into the level consistency quality control results. If the filtering result is empty, the judgment condition is read from the level judgment condition data, and based on the condition comparison record and the comparison result, it is judged whether each judgment condition forms a complete set of corresponding comparison results with any text condition group. The judgment condition that does not form a complete set of corresponding comparison results is determined as a level basis insufficient judgment condition, a level basis insufficient mark is generated, and the level basis insufficient judgment condition is written into the level consistency quality control result. If the supplementary description data includes at least one supplementary description text condition group, a supplementary description prompt mark is generated, and the supplementary description text condition group is written into the level consistency quality control result.

7. The method for risk control of early warning content based on multi-granularity semantic analysis according to claim 6, characterized in that, The method for generating supplementary description prompts includes: Read the supplementary description text condition group, condition comparison record, judgment condition for forming the condition comparison record, corresponding condition content type and non-corresponding condition content type from the supplementary description data; Write the supplementary description text condition group, the judgment condition, the condition comparison record, the corresponding condition content type and the non-corresponding condition content type into the supplementary description text condition group field, the judgment condition field, the condition comparison record field, the corresponding type field and the non-corresponding type field respectively to obtain the quality control prompt content; Generate supplementary description prompts and write the quality control prompts into the level consistency quality control results.

8. The method for risk control of early warning content based on multi-granularity semantic analysis according to claim 4, characterized in that, The method for comparing the numerical judgment content in the text condition group record and the judgment condition record includes: When the condition content type involved in the comparison is numerical judgment content, the numerical limitation direction, limitation value, and unit of measurement are extracted from the numerical judgment content in the main text condition group record and the judgment condition record, respectively. When the unit of measurement in the main text condition group record is inconsistent with the unit of measurement in the judgment condition record, the units of the limited values ​​in the main text condition group record and the limited values ​​in the judgment condition record are unified according to the pre-configured unit conversion rules. If the unit of measurement in the main text condition group record is consistent with the unit of measurement in the judgment condition record, or after the unit is unified, the corresponding judgment is made on the numerical limitation direction and limitation value in the main text condition group record and the numerical limitation direction and limitation value in the judgment condition record according to the preset numerical judgment correspondence rules, and the numerical judgment content judgment result is obtained. When the result of the numerical determination is that a correspondence is formed, a corresponding mark is written into the condition comparison record; when the result of the numerical determination is that a correspondence is not formed, a non-corresponding mark is written into the condition comparison record.

9. The method for risk control of early warning content based on multi-granularity semantic analysis according to claim 1, characterized in that, The method for obtaining the level determination condition data includes: Search the warning level specification data for a level determination statement that matches the warning type and the warning level. The level determination statement is subjected to text normalization processing, and the normalized level determination statement is subjected to condition content extraction to obtain the condition content extraction result. The content type of the condition content extraction result is determined from the indicator object, time limit, numerical determination content and occurrence status. Identify the common-factual expressions and alternative-factual expressions in the level judgment statement, and combine or segment the condition content extraction results based on the common-factual expressions and alternative-factual expressions to obtain the judgment conditions; Assign a judgment condition identifier to the judgment condition, organize the judgment condition identifier, the indicator object, time limit, numerical judgment content and occurrence status in the judgment condition into a judgment condition record, and summarize the judgment condition record to obtain the level judgment condition data.

10. A risk control system for early warning content based on multi-granularity semantic analysis, used to implement the risk control method for early warning content based on multi-granularity semantic analysis as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire the quality control early warning data and the early warning level specification data. The quality control early warning data includes the early warning type, early warning level and early warning text. The level determination condition forming module is used to combine the warning type and the warning level, extract the determination conditions corresponding to the warning type and the warning level from the warning level specification data, and obtain the level determination condition data. The text condition group forming module is used to identify the condition content in the warning text and group the condition content that represents the same condition to obtain text condition group data. The level support data forming module is used to identify the text condition groups in the text condition group data that support the warning level based on the level determination condition data, and to obtain level support data; The supplementary description data forming module is used to identify supplementary descriptions for text condition groups that have not formed level support based on the level support data, and obtain supplementary description data. The quality control result generation module is used to combine the level support data, the supplementary description data, and the level judgment condition data to perform quality control on the level consistency between the warning level and the warning text, and obtain the level consistency quality control result.

Images