An index system automatic decomposition and naming method based on semantic parsing

Abstract

This invention relates to the field of data governance technology, specifically to an automatic decomposition and naming method for an indicator system based on semantic parsing, comprising the following steps: S1: Obtain the indicator processing caliber text and generate structured indicator elements; S2: Establish an indicator reference link; S3: Map the structured indicator elements to a word root sequence and concatenate them to generate a preliminary English name; S4: Determine whether the preliminary English name needs abbreviation and output the English name to be verified; S5: Perform semantic deduplication and word root consistency verification on the English name to be verified, and output the final standardized indicator name after passing the verification. This invention converts the indicator processing caliber into structured indicator elements through semantic parsing, and combines this with an indicator reference link, a hierarchical word root library, constrained abbreviation generation, and deduplication verification mechanisms to automatically generate standardized English indicator names, thereby improving the consistency, accuracy, and maintenance efficiency of indicator system naming.

Description

Technical Field

[0001] This invention relates to the field of data governance technology, and in particular to an automatic decomposition and naming method for an indicator system based on semantic parsing. Background Technology

[0002] With the development of enterprise data governance, business analysis, and indicator platform construction, a large number of indicator processing definitions have been generated in various business systems. Indicator processing definitions typically include the measurement object, aggregation function, statistical time period, statistical conditions, grouping dimensions, and references to existing indicators. To facilitate indicator retrieval, data development, API calls, and report display, it is necessary to convert these indicator processing definitions into unified, standardized, and reusable English indicator names. Current indicator naming methods often rely on manual translation, abbreviation, and naming after reading the indicator definitions, or simply concatenating field names, function names, and dimension names using simple templates. This makes it difficult to systematically express the reference relationships, derivation relationships, and dimension inheritance relationships within complex indicators.

[0003] Existing technologies for naming indicators suffer from the following main problems: First, insufficient semantic parsing of the indicator processing text makes it difficult to accurately extract the relationships between measurement objects, aggregation functions, dimension constraints, and referenced indicators. Second, the lack of classification processing for basic, combined, and derived indicators makes it easy for indicators with different structures to use the same naming logic. Third, the lack of a unified root word library constraint in the English name generation process leads to different English expressions for the same business meaning in different indicators. Fourth, when names are long, direct manual abbreviation or free abbreviation generation by the model can easily cause missing core roots, abbreviation conflicts, and inconsistent naming styles. Therefore, it is necessary to provide an automatic decomposition and naming method for indicator systems based on semantic parsing. Through structured parsing, indicator reference link establishment, hierarchical root word mapping, constrained abbreviation generation, and semantic deduplication verification, the method can achieve automated, standardized, and consistent generation of English names for indicators. Summary of the Invention

[0004] To achieve the above objectives, this invention provides an automatic decomposition and naming method for an indicator system based on semantic parsing.

[0005] An automatic decomposition and naming method for an indicator system based on semantic parsing includes the following steps: S1: Obtain the text of the indicator processing criteria, extract the measurement object, aggregation function, dimension limitation and the identifier of the referenced indicator based on semantic parsing, and generate structured indicator elements; S2: Based on the structured indicator elements, identify the dependencies between the measurement object, aggregation function, dimension constraints, and the referenced indicator identifier, classify the indicators to be named into basic indicators, combined indicators, or derived indicators, and establish indicator reference links. S3: Based on a pre-built hierarchical root word library, structured indicator elements are mapped to root word sequences, and combined with the indicator type and indicator reference link of the indicator to be named, a preliminary English name is generated by splicing them in a preset order. S4: Determine whether the initial English name needs to be abbreviated based on the preset length threshold; if abbreviation is required, input the initial English name, historical indicator naming corpus, and industry abbreviation database into the preset large language model to generate candidate abbreviation names, and select the recommended abbreviation name as the English name to be verified based on the consistency score; if abbreviation is not required, use the initial English name as the English name to be verified. S5: Perform semantic deduplication and root consistency verification on the English name to be verified. If it passes, output the final standardized index name and update the newly generated root mapping relationship to the hierarchical root library.

[0006] Optionally, S1 specifically includes: S11: Perform word segmentation on the text of the indicator processing criteria to obtain a text word sequence; S12: Perform entity recognition and syntactic dependency analysis on the text word sequence, identify words representing statistical objects as measurement objects, and identify words representing statistical methods as aggregation functions; S13: Identify terms representing statistical range, statistical time, statistical conditions, and grouping fields as dimension constraints; S14: Identify existing indicator names, indicator codes, or indicator aliases appearing in the indicator processing caliber text as the identifiers of the referenced indicators; S15: Generate structured indicator elements based on the measurement object, aggregation function, dimension limitation, and referenced indicator identifier.

[0007] Optionally, the classification of the indicators to be named into basic indicators, combined indicators, or derived indicators in step S2 specifically includes: When the structured indicator elements do not contain the identifier of the referenced indicator, but contain the measurement object and the aggregation function, the indicator to be named will be classified as a basic indicator. When a structured indicator element contains at least two referenced indicator identifiers, and there is a computational relationship between the referenced indicator identifiers, the indicator to be named will be classified as a combined indicator. When a structured indicator element contains the identifier of a referenced indicator and includes time offset, ratio conversion, difference calculation, or month-on-month calculation relationship, the indicator to be named will be classified as a derived indicator.

[0008] Optionally, establishing the indicator reference link in S2 specifically includes: Based on the operational, derivational, and dimensional inheritance relationships between the referenced metric identifier and the unnamed metric, a metric reference chain is generated. The metric reference chain includes the current metric identifier, the referenced metric identifier, the reference relationship type, and the reference order. The reference relationship type includes operational reference, derivational reference, and dimensional inheritance reference.

[0009] Optionally, the hierarchical root word library includes an object root word layer, a function root word layer, a dimension root word layer, a relation root word layer, and an abbreviation root word layer.

[0010] Optionally, S3 specifically includes: S31: Based on the hierarchical root word library, the measurement objects in the structured indicator elements are mapped to object roots, the aggregation functions are mapped to function roots, and the dimension constraints are mapped to dimension roots; S32: Extract the corresponding type root from the relation root layer according to the indicator type of the indicator to be named; S33: When the indicator to be named has an indicator reference link, extract the corresponding relation root from the relation root layer according to the reference relationship type; S34: Concatenate the root words in the order of dimension, object, function, type, and relation to generate a preliminary English name.

[0011] Optionally, the preset length threshold in S4 includes a character length threshold of 30 and a word root number threshold of 5; If the number of characters in the preliminary English name is greater than 30, or the number of word roots in the preliminary English name is greater than 5, the preliminary English name is determined to need to be abbreviated. If the number of characters in the preliminary English name is no more than 30 and the number of word roots in the preliminary English name is no more than 5, it is determined that the preliminary English name does not need to be abbreviated.

[0012] Optionally, the step of inputting the preliminary English name, historical indicator naming corpus, and industry abbreviation database into a preset large language model to generate candidate abbreviation names specifically includes: S41: Perform root word splitting on the preliminary English name to obtain the root word sequence to be abbreviated, and extract historical English names with the same indicator type or the same dimension root words as the root word sequence to be abbreviated from the historical indicator naming corpus to form a historical naming reference set. S42: Extract standard abbreviation roots from the industry abbreviation database that correspond to each root in the sequence of root words to be abbreviated, and form an abbreviation constraint set; S43: Generate model input prompt information based on the sequence of word roots to be abbreviated, the historical naming reference set, and the abbreviation constraint set. The model input prompt information includes core word roots that must be retained, non-core word roots that are allowed to be abbreviated, word roots that are prohibited from being replaced, and the number of candidate names. S44: Input the model input prompt information into the preset large language model, so that the preset large language model performs abbreviation and recombination of the initial English name according to the abbreviation constraint set, and outputs multiple candidate abbreviation names.

[0013] Optionally, the step S4, which selects the recommended abbreviation as the English name to be verified based on the consistency score, specifically includes: Calculate the root word retention score, historical pattern matching score, and abbreviation conflict deduction value for each candidate abbreviation name; A consistency score is generated based on the root word retention score, the historical pattern matching score, and the abbreviation conflict deduction value. The candidate abbreviation with the highest consistency score is selected as the recommended abbreviation.

[0014] Optionally, S5 specifically includes: S51: Perform semantic similarity matching between the English name to be verified and the stored standardized indicator names; When the semantic similarity reaches the preset deduplication threshold, the corresponding stored standardized indicator name is output. When the semantic similarity does not reach the preset deduplication threshold, the mapping relationship between the word roots in the English name to be verified and the structured index elements is verified by word root consistency. S52: When the root word consistency check passes, output the English name to be checked as the final standardized indicator name; S53: Write the root mapping relationships of the English names to be verified that are not recorded in the hierarchical root database into the corresponding root layer.

[0015] The beneficial effects of this invention are: This invention performs semantic parsing on the text of indicator processing specifications to extract the measurement object, aggregation function, dimension limitation, and the identifier of the referenced indicator. It further identifies the indicator type and establishes the indicator reference link, so that indicator naming no longer depends on human understanding of the text of specifications, but is automatically completed by decomposition and naming based on structured indicator elements. As a result, it can accurately distinguish between basic indicators, combined indicators, and derived indicators, and generate English names for indicators with different composition methods according to corresponding rules, thereby improving the accuracy and traceability of indicator naming.

[0016] This invention maps structured indicator elements to unified root words through a hierarchical root word library. When the initial English name exceeds the length threshold, it generates candidate abbreviations by combining historical indicator naming corpora and industry abbreviation libraries. Finally, it outputs the standardized indicator name through consistency scoring, semantic deduplication, and root word consistency verification. This avoids multiple English expressions for the same business meaning, reduces abbreviation conflicts and duplicate naming, and ensures that indicator names remain consistent, concise, and standardized in data development, indicator retrieval, interface calls, and report display. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the automatic decomposition and naming method of the indicator system according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the process of generating a preliminary English name according to an embodiment of the present invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should also be noted that, to make the embodiments more comprehensive, the following embodiments are the best and preferred embodiments, and those skilled in the art can use other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0020] It should be noted that the use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.

[0021] Generally, terms can be understood at least partly from their use in context. For example, depending at least partly on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in a singular sense, or a combination of features, structures, or characteristics in a plural sense. Additionally, the term "based on" can be understood not necessarily to convey an exclusive set of factors, but rather, alternatively, depending at least partly on the context, to allow for the presence of other factors that are not necessarily explicitly described.

[0022] like Figures 1-2 As shown, an automatic decomposition and naming method for an indicator system based on semantic parsing includes the following steps: S1: Obtain the text of the indicator processing criteria, extract the measurement object, aggregation function, dimension limitation and the identifier of the referenced indicator based on semantic parsing, and generate structured indicator elements; S1 specifically includes: S11: Perform word segmentation on the text of the indicator processing criteria to obtain a text word sequence; S12: Perform entity recognition and syntactic dependency analysis on the text word sequence, identify words representing statistical objects as measurement objects, and identify words representing statistical methods as aggregation functions; S13: Identify terms representing statistical range, statistical time, statistical conditions, and grouping fields as dimension constraints; S14: Identify existing indicator names, indicator codes, or indicator aliases appearing in the indicator processing caliber text as the identifiers of the referenced indicators; S15: Generate structured indicator elements based on the measurement object, aggregation function, dimension limitation, and referenced indicator identifier.

[0023] Specifically, after obtaining the text of the indicator processing guidelines corresponding to the indicator to be named, the text is first standardized by converting full-width characters to half-width characters, unifying the case, removing duplicate spaces and invalid separators, and converting parentheses, colons, forward slashes, minus signs, plus signs, percent signs, and equal signs in the text into standard operation or delimiter marks. This ensures that indicator processing guidelines from different sources form a unified parsing format, improving the stability of subsequent recognition results. Next, the standardized indicator processing guidelines text is segmented to obtain a text word sequence. During word segmentation, the preset business field dictionary, indicator name dictionary, and industry abbreviation dictionary are first used for longest word matching, retaining consecutive field names, indicator names, and abbreviations with complete business meaning as whole words. Then, the remaining text is segmented according to semantic boundaries, operators, and conjunctions to form a text word sequence arranged in the original text order. Each word record its content, type, start character position, and end character position. This process avoids breaking down words with overall meaning, such as order amount, number of active users, and year-on-year growth rate, into irrelevant characters, thus providing a stable lexical basis for subsequent semantic recognition.

[0024] Furthermore, the text lexical sequence is used to identify the measurement objects. Specifically, noun lexical elements, business field lexical elements, and data object lexical elements in the lexical sequence are matched with a pre-defined measurement object dictionary to identify data items representing statistical targets as measurement objects. When both statistical action words and business object words exist in the same sentence, business object words that have an object relationship or modification relationship with the statistical action words are identified as measurement objects. When words such as amount, quantity, number of people, frequency, order, income, cost, duration, area, number of items, inventory, and sales volume appear in the text, representing statistically significant objects, they are classified into the measurement object field. Through this processing, the core objects statistically represented by the indicator to be named can be clearly identified, avoiding inaccurate names generated solely based on generalized business terms in subsequent naming.

[0025] Furthermore, aggregation functions are identified in the text word sequence. Specifically, verbal statistical words, mathematical operation words, and statistical function words in the word sequence are matched with a pre-defined aggregation function dictionary, and words representing statistical methods are identified as aggregation functions. Among them, summation, total, cumulative, and total are mapped to summation functions; count, quantity, number of transactions, and frequency are mapped to count functions; deduplication, independent, and unique are mapped to deduplication count functions; average, mean, per capita, and single average are mapped to average functions; maximum, highest, and peak are mapped to maximum value functions; and minimum, lowest, and valley are mapped to minimum value functions. For cases where statistical function words do not appear directly in the text but compound words such as total sales, number of orders, and average order value exist, the corresponding aggregation function is determined based on the prefixes or suffixes in the compound words. Through this processing, different expressions can be unified into a unified aggregation function type, so that subsequent indicator classification and English word root mapping have a consistent basis.

[0026] Furthermore, dimensional constraint identification is performed on the text lexical sequence. Specifically, based on the semantic category and syntactic modification relationship of the lexical, lexical that limits the statistical scope, statistical time, statistical conditions, and grouping fields are categorized into dimension constraint fields. For text containing time-related lexical such as "today," "this week," "this month," "this quarter," "this year," "last 7 days," "last 30 days," "year-on-year period," and "month-on-month period," the corresponding lexical is identified as a statistical time constraint. For text containing category fields such as "region," "store," "channel," "category," "customer level," "department," "project," and "equipment type," the corresponding lexical is identified as a grouping field constraint. For text containing conditional lexical such as "greater than," "less than," "equal to," "not empty," "paid," "completed," "valid," "excluded," "only counted," and "filtered," the corresponding lexical and its associated fields are identified as statistical condition constraints. For text containing business boundary lexical, it is identified as a statistical scope constraint. This processing allows for the separate storage of constraints with different functions, avoiding confusion between time constraints, condition constraints, and grouping constraints, and facilitating the subsequent generation of corresponding dimension lexical roots.

[0027] Furthermore, the system identifies referenced indicator identifiers for the text lexical sequences. Specifically, the text lexical sequences are matched against an existing indicator library. If the lexical content matches an existing indicator name, the lexical is identified as a referenced indicator identifier. If the lexical content conforms to a preset indicator coding format, the indicator code is matched against the indicator coding field in the indicator library, and the matching result is identified as a referenced indicator identifier. If the lexical content matches an alias in a preset indicator alias mapping table, the corresponding referenced indicator identifier is determined based on the alias mapping relationship. For expressions in the text containing forward slashes, plus signs, minus signs, multiplication signs, division signs, equal signs, and parentheses, the system further identifies existing indicator names, indicator codes, or indicator aliases appearing on both sides of the expression or within the parentheses, and records them as a set of referenced indicator identifiers according to their order of appearance. Through this processing, the source of reference between the indicator to be named and existing indicators can be clearly identified, providing a foundation for establishing indicator reference links in the future.

[0028] Furthermore, the correspondences between the measurement object, aggregation function, dimension constraints, and referenced indicator identifiers are organized to generate structured indicator elements. Specifically, based on syntactic dependency relations and lexical positional relations, aggregation functions are bound to the measurement objects they operate on, and dimension constraints are bound to the measurement objects they constrain or the referenced indicator identifiers they are referencing. The referenced indicator identifiers are recorded according to the order of operations in the text. Structured indicator elements include at least a measurement object field, an aggregation function field, a dimension constraint field, and a referenced indicator identifier field. The measurement object field records the statistical objective, the aggregation function field records the statistical method, the dimension constraint field records the statistical range, statistical time, statistical conditions, and grouping fields, and the referenced indicator identifier field records the existing indicators being invoked. Through this processing, the indicator processing caliber text can be converted into structured data with clearly defined fields and relationships, providing a unified input for subsequent indicator type determination, indicator reference link establishment, and English name generation.

[0029] Structured indicator elements include the measurement object field, aggregation function field, dimension limitation field, referenced indicator identifier field, and element source location field.

[0030] In one specific embodiment, the text for processing the indicator is: Sum the order amounts of all paid orders in the past 30 days, group them by channel, and use the GMV indicator for validation. After parsing this text, the following structured indicator elements are obtained: In the example above, order amount is used as a measurement object field to represent the statistical object of the indicator to be named; summation is used as an aggregation function field to represent the summation of order amount; the last 30 days, paid orders, and channels are written into the dimension limitation field as statistical time, statistical conditions, and grouping fields, respectively; GMV indicator is used as a referenced indicator identifier field to indicate that there is a reference to an existing indicator in the indicator processing document; through the above fielded records, the indicator processing document in natural language form can be converted into structured indicator elements that can be directly called in subsequent steps.

[0031] S2: Based on the structured indicator elements, identify the dependencies between the measurement object, aggregation function, dimension constraints, and the referenced indicator identifier, classify the indicators to be named into basic indicators, combined indicators, or derived indicators, and establish indicator reference links. In S2, the indicators to be named are categorized into basic indicators, combined indicators, or derived indicators, specifically including: When the structured indicator elements do not contain the identifier of the referenced indicator, but contain the measurement object and the aggregation function, the indicator to be named will be classified as a basic indicator. When a structured indicator element contains at least two referenced indicator identifiers, and there is a computational relationship between the referenced indicator identifiers, the indicator to be named will be classified as a combined indicator. When a structured indicator element contains the identifier of a referenced indicator and includes time offset, ratio conversion, difference calculation, or month-on-month calculation relationship, the indicator to be named will be classified as a derived indicator.

[0032] Specifically, when the structured indicator elements do not contain the identifier of the referenced indicator, but contain the measurement object and the aggregation function, the indicator to be named is classified as a basic indicator. Among them, the basic indicator is an indicator formed by direct statistics of the original business fields, and its naming is mainly based on the measurement object, the aggregation function and the dimension limitation. Through this classification, direct statistical indicators and referenced indicators can be distinguished, and the reference relationship root words are not mistakenly added when generating names later.

[0033] When a structured indicator element contains at least two referenced indicator identifiers, and there are addition, subtraction, multiplication, division, bracket combinations, or weighted combinations among these referenced indicator identifiers, the indicator to be named is classified as a combined indicator. A combined indicator is an indicator formed by combining multiple existing indicators through operational relationships. Its reference source and operational order can be used to establish the indicator reference chain in the future. Through this classification, the combination relationship between multiple referenced indicators can be clarified, so that subsequent naming can reflect the source structure of the combined indicator.

[0034] When a structured indicator element contains the identifier of a referenced indicator and includes time offset, proportional conversion, difference calculation, or month-on-month calculation relationship, the indicator to be named is classified as a derived indicator. Derived indicators are those formed by secondary calculations based on existing indicators. Time offset includes the time correspondence between the current period and the previous period, and the same period last year; proportional conversion includes calculations of percentage, conversion rate, and growth rate; difference calculation includes the difference between the current value and the benchmark value; and month-on-month calculation includes the change between the current statistical period and the previous statistical period. This classification distinguishes derived calculation indicators from indicators formed by direct combinations of multiple indicators, ensuring that subsequent naming accurately utilizes the derived relationship terminology.

[0035] When the same unnamed indicator simultaneously meets some of the conditions of both combined and derived indicators, priority determination is performed in the order of derived indicators, combined indicators, and basic indicators. That is, first, it is determined whether there is a relationship of time offset, ratio conversion, difference calculation, or month-on-month calculation. If so, it is classified as a derived indicator. If there is no such relationship but there are at least two referenced indicator identifiers and their operation relationship, it is classified as a combined indicator. If there is no referenced indicator identifier but there is a measurement object and an aggregation function, it is classified as a basic indicator. Through this priority determination, the same unnamed indicator can correspond to only one indicator type, avoiding classification result conflicts.

[0036] In a specific example, the metric processing text is: "Sum of the order amounts of all paid orders this month." Parsing reveals the metric object as order amount, the aggregation function as summation, the dimensions as including this month and paid orders, and the referenced metric identifier as empty; therefore, it is categorized as a basic metric.

[0037] In another specific example, the indicator processing text is: Sales Profit Margin = Sales Profit / Sales Revenue. Parsing reveals that the referenced indicator includes both sales profit and sales revenue, and there is a division relationship between the two. Therefore, it is classified as a combined indicator.

[0038] In another specific example, the indicator processing text is: the month-on-month growth rate of this month's sales revenue relative to last month's sales revenue. Parsing reveals that the referenced indicator is identified as sales revenue, and the corresponding time offset and month-on-month calculation relationship for this month, last month, and the growth rate are identified. Therefore, it is classified as a derived indicator. In summary, by determining the type of the indicator to be named through the field content and relationships in the structured indicator elements, the basic indicators, combined indicators, and derived indicators have a clear basis for classification. The classification results are then passed to the subsequent root word mapping and English name generation process, thereby improving the consistency of automatic decomposition and naming of the indicator system.

[0039] Establishing an indicator referencing link in S2 specifically includes: Based on the operational, derivational, and dimensional inheritance relationships between the referenced metric identifier and the metric to be named, a metric reference chain is generated. The metric reference chain includes the current metric identifier, the referenced metric identifier, the reference relationship type, and the reference order. Among them, the reference relationship types include operational reference, derivational reference, and dimensional inheritance reference.

[0040] Establishing an indicator reference link in S2 refers to constructing link data representing the dependency path between indicators based on the operational, derivational, and dimensional inheritance relationships between the referenced indicator identifier and the indicator to be named, after generating structured indicator elements and determining indicator types. The indicator reference link includes at least the current indicator identifier, the referenced indicator identifier, the reference relationship type, and the reference order. This process makes the dependency relationship between the indicator to be named and existing indicators explicit, providing a basis for generating preliminary English names by combining the reference link. Specifically, the referenced indicator identifier field in the structured indicator elements is first read, and each referenced indicator identifier is matched with existing indicator records in the indicator library to obtain the corresponding unique identifier of the referenced indicator. For referenced indicator identifiers that do not match a unique record, they are recorded as reference identifiers to be confirmed and do not participate in the final link output; for referenced indicator identifiers that match successfully, they are written as link nodes into the candidate reference set.

[0041] Furthermore, the system identifies operational reference relationships based on the operators and semantics of the operations in the indicator processing text. When there are expressions such as plus, minus, multiplication, division, parentheses, weighting, percentage, or difference between the referenced indicator identifiers, the relationship between the corresponding referenced indicator and the current indicator is marked as an operational reference, and the operator and its order in the expression are recorded. For the expression Sales Profit Margin = Sales Profit / Sales Revenue, the system uses both Sales Profit and Sales Revenue as referenced indicator identifiers and records the relationship type between them and the current indicator as an operational reference. Through this processing, the source of the combined indicators' operations can be preserved, allowing subsequent naming to reflect the indicator composition relationship.

[0042] Furthermore, derived reference relationships are identified based on time offset, proportional conversion, difference calculation, and month-on-month calculation semantics. When derived calculation terms such as "current period," "previous period," "same period," "last month," "last year's same period," "month-on-month," "year-on-year," "growth rate," "change rate," "difference," or "benchmark value" appear in the indicator processing text, the relationship between the corresponding referenced indicator and the current indicator is marked as a derived reference, and the type of derived relationship is recorded. For the month-on-month growth rate of this month's sales relative to last month's sales, the system uses sales as the referenced indicator identifier and records the relationship type between it and the current indicator as a derived reference. Through this processing, secondary calculation relationships based on existing indicators are distinguished from ordinary operational combination relationships, enabling subsequent naming to call the corresponding derived relationship root words. Furthermore, dimension inheritance relationships are identified based on dimension limitation fields. When the indicator to be named uses the statistical range, statistical time, statistical conditions, or grouping fields of the referenced indicator, the relationship between the corresponding referenced indicator and the current indicator is marked as a dimension inheritance reference; when the indicator to be named adds or replaces dimension limitations based on the referenced indicator, the inherited dimension and the changed dimension are recorded. For the GMV metric calculated by channel over the past 30 days, the system identifies that the current metric references the GMV metric and inherits its statistical scope. It also adds the past 30 days and channel dimension limitations, thus recording this reference relationship as a dimension inheritance reference. This process preserves the source of dimension inheritance in the metric's scope, preventing the omission of key limiting information during subsequent name generation.

[0043] After determining the type of reference relationship, the reference relationship strength is calculated for each reference relationship in the candidate reference set. This strength is used to determine the primary relationship type when multiple relationships exist for the same referenced indicator. The reference relationship strength can be calculated using the following expression: ,in, Indicates the first The strength of the referencing relationship between the referenced metric and the current metric; This value represents the matching value of the operation relationship. It is 1 when an operator or operation semantic exists, and 0 otherwise. This represents the derived relation matching value, which is 1 when there is semantics such as time offset, proportional conversion, difference calculation or month-on-month calculation, and 0 otherwise. This indicates the dimension inheritance matching value, which is 1 when there is an inheritance relationship in the statistical range, statistical time, statistical conditions, or grouping fields, and 0 otherwise. Let represent the weights of the operational relationship, the derivation relationship, and the dimension inheritance relationship, respectively, with values ​​of 0.4, 0.4, and 0.2, and satisfying the following conditions: This expression can transform different reference relationships into comparable strength values, giving the reference link construction process a definite basis for judgment.

[0044] When the same referenced indicator satisfies multiple reference relationships simultaneously, the primary reference relationship type is determined in the order of derived reference, operational reference, and dimension inheritance reference. If there is no derived reference but an operational reference exists, the primary reference relationship type is operational reference. If only a dimension inheritance relationship exists, the primary reference relationship type is dimension inheritance reference. Simultaneously, the system retains non-primary relationships as auxiliary relationship markers. This process ensures the uniqueness of the link type while preserving complete relationship information, enabling subsequent naming to determine both the primary root and necessary qualifying roots. Furthermore, the reference order is determined based on the position of the referenced indicator identifier in the indicator processing document, the expression operation priority, and the direction of derivation calculation. The reference order can be calculated using the following expression: ,in, Indicates the first The reference order value of each referenced indicator; the smaller the value, the earlier the reference order. Indicates the first The normalized position number of each cited indicator in the indicator processing caliber text; Indicates the first The priority number of the referenced index in the expression operation is 1 for operations within parentheses, 2 for multiplication and division, 3 for addition and subtraction, and 4 for no explicit operation. This indicates the direction number of the derived calculation, with the base period indicator taking 1, the current period indicator taking 2, and the non-derived indicator taking 3. These represent the weights for positional order, computational priority, and derivation direction, respectively, with values ​​of 0.3, 0.4, and 0.3, and satisfying the following conditions: This expression allows for the determination of a stable citation order when multiple referenced indicators exist, providing a sequential basis for subsequent root word concatenation and name generation. Finally, indicator citation link records are generated based on the current indicator identifier, the referenced indicator identifier, the citation relationship type, the auxiliary relationship marker, the citation order, and the relationship strength. When the indicator to be named references multiple existing indicators, these link records are combined into an indicator citation link set according to the citation order. When the indicator to be named does not contain a referenced indicator identifier, the indicator citation link is set to empty, and this indicator is used as the base indicator in the subsequent root word mapping process. This process ensures that different types of indicators to be named have clear subsequent processing paths.

[0045] In a specific example, the indicator processing text is: Sales Profit Margin = Sales Profit / Sales Revenue. The current indicator is identified as Sales Profit Margin. The referenced indicators include Sales Profit and Sales Revenue, which have a division relationship. Therefore, both reference relationships are marked as operational references. Since Sales Profit is on the left side of the division sign and acts as the dividend, while Sales Revenue is on the right side and acts as the divisor, the system determines the reference order of Sales Profit as 1 and the reference order of Sales Revenue as 2. This generates the following indicator reference chain: the current indicator Sales Profit Margin points to both the referenced indicators Sales Profit and Sales Revenue, with both reference relationships being operational references, and the reference order being 1 and 2 respectively.

[0046] S3: Based on a pre-built hierarchical root word library, structured indicator elements are mapped to root word sequences, and combined with the indicator type and indicator reference link of the indicator to be named, a preliminary English name is generated by splicing them in a preset order. The hierarchical root word library includes object root word layer, function root word layer, dimension root word layer, relation root word layer and abbreviation root word layer.

[0047] The hierarchical root word library structure is as follows: This process extracts existing Chinese indicator names, existing English names, indicator processing guidelines, and manually confirmed indicator types from historical indicator naming corpora. It then aligns the business objects, statistical methods, dimensional limitations, and reference relationships appearing in the Chinese indicator names and indicator processing guidelines with the English words, abbreviations, and conjunctions in the existing English names. After alignment, the English expressions corresponding to the same Chinese semantic element are merged, retaining the most frequent and manually confirmed English expressions as standard root words. When the same Chinese semantic element corresponds to multiple English expressions, the non-standard English expressions are recorded as alternative root words.

[0048] Furthermore, based on the semantic function of the root words, the basic mapping relationships are divided into an object root word layer, a function root word layer, a dimension root word layer, a relation root word layer, and an abbreviation root word layer. Specifically, the object root word layer stores the mapping relationship between metric objects and English root words; the function root word layer stores the mapping relationship between aggregate functions and English root words; the dimension root word layer stores the mapping relationship between dimension constraints and English root words; the relation root word layer stores the mapping relationship between indicator types and reference relationship types and English connective root words; and the abbreviation root word layer stores the mapping relationship between confirmed abbreviation names and full English root words. This hierarchical storage allows roots with different semantic functions to be called separately when generating names, avoiding the mixing of object, function, and relation root words. The object root word layer includes at least the object's Chinese name, the object's standard English root word, the object's alias root word, the object's business domain, and its enabled status. The object's Chinese name records the Chinese expression in the metric object's fields. The object's standard English root word records the standard English expression corresponding to the metric object. The object's alias root word records the English names that appeared in historical corpora but were not used as standard expressions. The object's business domain distinguishes the meaning of the same Chinese word in different business scenarios. The enabled status indicates whether the root word participates in the current naming process. The function root word layer includes at least the function's Chinese name, standard function identifier, function's English root word, and applicable statistical method. The function's Chinese name records Chinese statistical expressions such as summation, counting, averaging, deduplication counting, maximum value, and minimum value. The standard function identifier records the corresponding standard function type. The function's English root word records the function expression in the English name. The applicable statistical method limits the statistical meaning corresponding to the function root word. The dimension root word layer includes at least the dimension's Chinese name, dimension category, dimension's standard English root word, and dimension value format. The dimension category includes statistical time, statistical range, statistical conditions, and grouping fields. The dimension's standard English root word records the dimension's expression in its English name. The dimension value format constrains the expression of dimensions such as time, status, channel, region, and category in their English names. This layer allows for mapping dimensions such as the past 30 days, paid, channel, and region separately according to their semantic categories, avoiding mixed representations of different dimension types in the names. The relation root layer includes at least the indicator type, the reference relationship type, the relation root in English, and the concatenation position. Indicator types include basic indicators, combined indicators, and derived indicators; reference relationship types include operational references, derived references, and dimension inheritance references; the relation root in English is used to represent the connection expression of indicator types or reference relationships in English names; the concatenation position is used to specify the insertion position of the relation root in the root sequence. The abbreviation root layer includes at least the complete English root, the standard abbreviation root, the source of the abbreviation, the applicable business domain, and conflict markers.Full English root words are used to record unabbreviated English expressions. Standard abbreviation root words are used to record abbreviation expressions confirmed through historical naming corpora or industry abbreviation thesaurus. Abbreviation source is used to record that the abbreviation comes from historical indicator naming corpora or industry abbreviation thesaurus. Applicable business domain is used to limit the scope of use of the abbreviation. Conflict flag is used to identify whether the same abbreviation corresponds to multiple full English root words.

[0049] For ease of implementation, the hierarchical root word library can be stored using the following structure: After construction, a root number can be generated for each root mapping relationship, and the creation source, update time, and review status can be recorded. Review status includes pending review, enabled, and disabled; only root mapping relationships in the enabled state participate in subsequent root sequence generation. For newly appearing and verified root mapping relationships in S5, the system writes them into the corresponding root layer according to their semantic function and marks the review status as enabled or pending review. This process ensures that the hierarchical root library is continuously updated along with the indicator naming process, while guaranteeing that the root mapping relationships participating in automatic naming have a controllable source.

[0050] In a specific example, when the historical metric naming corpus contains the English root "orderamt" for order amount, "sum" for summation, "last 30d" for the past 30 days, "chnl" for channel, and "mom growth" for month-on-month growth rate, these mappings are written into the object root layer, function root layer, dimension root layer, and relation root layer, respectively. When the industry abbreviation corpus contains the abbreviation "amt" for amount and "chnl" for channel, the system writes them into the abbreviation root layer. Subsequently, when naming the metric processing text that calculates the sum of order amounts from various channels over the past 30 days, "orderamt" can be called from the object root layer, "sum" from the function root layer, and "last 30d" and "chnl" from the dimension root layer, generating the corresponding root sequence according to preset concatenation rules. By standardizing and organizing historical naming corpora, business field dictionaries, aggregation function rule tables, dimension field dictionaries, indicator type rule tables, and industry abbreviation libraries, a hierarchical root library is constructed, including object root layers, function root layers, dimension root layers, relation root layers, and abbreviation root layers. This enables structured indicator elements, indicator types, and indicator reference links to be mapped to concatenable, verifiable, and updatable English root words, thus providing a unified root foundation for subsequent preliminary English name generation, candidate abbreviation name screening, and final standardized indicator name output.

[0051] S3 specifically includes: S31: Based on the hierarchical root word library, the measurement objects in the structured indicator elements are mapped to object roots, the aggregation functions are mapped to function roots, and the dimension constraints are mapped to dimension roots; S32: Extract the corresponding type root from the relation root layer according to the indicator type of the indicator to be named; S33: When the indicator to be named has an indicator reference link, extract the corresponding relation root from the relation root layer according to the reference relationship type; S34: Concatenate the root words in the order of dimension, object, function, type, and relation to generate a preliminary English name.

[0052] In a specific example, the text for the indicator processing definition is: sum the order amounts of orders paid in the past 30 days, grouped by channel.

[0053] After parsing by S1, the structured indicator elements are as follows: the measurement object is order amount, the aggregation function is summation, the dimensions include the statistical time of the past 30 days, the statistical condition of paid orders, and the grouping field "channel". The referenced indicator is empty. S2 determines that this indicator to be named does not reference any existing indicators and contains a measurement object and an aggregation function; therefore, it is classified as a basic indicator, and the indicator reference link is empty.

[0054] Subsequently, word root mapping is performed based on the hierarchical word root library: The measurement of order amount is mapped to the root word "order amt". The summation of aggregate functions is mapped to the function root "sum". The statistical timeframe of the last 30 days is mapped to the dimension root word "last 30d". The statistical condition for paid orders is mapped to the dimension root word "paid order". The grouping field channel is mapped to the dimension root chnl; The basic indicator type is mapped to the type root, which serves as an internal concatenation constraint and is not included in the final concatenation result.

[0055] Following the concatenation rules corresponding to the basic indicators, the dimension root, object root, and function root are concatenated sequentially to obtain the root sequence: last 30d, paid order, chnl, order amt, sum; thus generating the preliminary English name: last30d paid order chnl order amt sum; where last 30d represents the statistical period of the past 30 days, paid order represents the condition of paid orders, chnl represents the channel grouping field, order amt represents the order amount measurement object, and sum represents the summation statistical method. This preliminary English name can fully reflect the statistical time, filtering conditions, grouping fields, measurement objects, and aggregation functions in the original indicator processing document.

[0056] In another specific example, the text for the indicator processing is: the month-on-month growth rate of this month's sales revenue relative to last month's sales revenue. After parsing in S1, the referenced indicator is identified as sales revenue, with dimensions limited to this month and last month, and the time offset and growth relationship corresponding to the month-on-month growth rate are identified. In S2, it is determined that the indicator to be named is a derived indicator, and a derived reference link from sales revenue to the current indicator is established. The root word for sales revenue, sales amt, the root word for this month, prev mon, the root word for last month, and the root word for the month-on-month growth rate, mom growth, are called from the hierarchical root word library. The preliminary English name is generated according to the derived indicator concatenation rules: curr mon prev mon sales amt mom growth. This name can reflect the current period, the comparison period, the referenced indicator, and the derived growth relationship. In summary, this implementation method maps the measurement object, aggregation function, dimension limitation, indicator type, and reference relationship through the hierarchical root word library, and executes the corresponding concatenation rules according to different indicator types, so that the preliminary English name can be generated by the structured indicator elements, thereby improving the interpretability and consistency of the indicator's English naming.

[0057] S4: Determine whether the initial English name needs to be abbreviated based on the preset length threshold; if abbreviation is required, input the initial English name, historical indicator naming corpus, and industry abbreviation database into the preset large language model to generate candidate abbreviation names, and select the recommended abbreviation name as the English name to be verified based on the consistency score; if abbreviation is not required, use the initial English name as the English name to be verified. The preset length thresholds in S4 include a character length threshold of 30 and a word root number threshold of 5. If the number of characters in the preliminary English name is greater than 30, or the number of word roots in the preliminary English name is greater than 5, the preliminary English name is determined to need to be abbreviated. If the number of characters in the preliminary English name is no more than 30 and the number of word roots in the preliminary English name is no more than 5, it is determined that the preliminary English name does not need to be abbreviated.

[0058] Specifically, the character length threshold is obtained by statistically analyzing the confirmed standardized English names in the historical indicator naming corpus. The number of characters in each English name in the historical indicator naming corpus is counted, and names manually marked as too long, containing duplicate word roots, or not conforming to naming standards are removed. The English names that have been manually confirmed and are normally used in the indicator management system are retained as a sample set. Then, the concentrated distribution range of the number of characters in the English names in the sample set is calculated, and the boundary value that covers the main standard names without causing truncation in field display is determined as the character length threshold. In this embodiment, the character length threshold is set to 30. That is, when the initial number of characters in the English name is greater than 30, the name is considered to exceed the standard naming length and needs to be abbreviated. The reason for setting the character length threshold to 30 is that the number of characters in the confirmed standardized English names in the historical indicator naming corpus is mainly concentrated between 18 and 30. When the number of characters exceeds 30, the name is prone to forming an overly long identifier when displayed in the indicator directory, data table fields, interface parameters, and report column names, which will reduce the efficiency of manual verification and system retrieval. Therefore, setting 30 as the character length threshold allows for control of name length while preserving the main semantic information, enabling initial English names exceeding this length to enter the controlled abbreviation process.

[0059] The root word count threshold was obtained by statistically analyzing the number of root words in historical standardized English names. Historical English names were broken down using connectors as boundaries to obtain the number of root words for each name, and the distribution of root word counts in confirmed standardized English names was statistically analyzed. For basic indicators, root words typically include time or conditional root words, object root words, and function root words; for combined and derived indicators, root words also include relational or derived root words. Statistically, when the number of root words in the English name does not exceed 5, it can cover the semantics of the main indicator; when the number of root words exceeds 5, the name usually contains multiple dimension constraints, multiple object root words, or multiple relational root words, resulting in semantic redundancy and space for abbreviation. The reason for setting the root word count threshold to 5 is that a standardized indicator name usually needs to express the core content of statistical time, statistical conditions or grouping fields, measurement objects, aggregation functions, and indicator relationships; five root words can cover the above main semantic units; when the number of root words exceeds 5, it indicates that the initial English name already contains a lot of limiting information or referencing relationship information, and continuing to retain it completely would lead to a verbose name. Using 5 as the threshold for the number of word roots can trigger abbreviation processing without weakening the core semantic expression, thus maintaining a balance between standardization and readability in the initial English name.

[0060] In this implementation, when the initial English name is determined to require abbreviation, it is not directly input into a preset large language model for free generation. Instead, the initial English name is first split into its root words, and constrained model input prompts are constructed by combining historical indicator naming corpora and industry abbreviation databases. Then, the preset large language model outputs multiple candidate abbreviation names. This process ensures that the candidate abbreviation names retain the core semantics of the initial English name while conforming to historical indicator naming habits and industry abbreviation rules. Specifically, the initial English name is first split into its root words to obtain a sequence of root words to be abbreviated. The root word splitting uses the underscore in the initial English name as the main boundary and verifies the splitting results by combining the object root word layer, function root word layer, dimension root word layer, and relation root word layer in the hierarchical root word database. For root words composed of compound abbreviations, the system determines whether to retain them as whole root words based on the abbreviation root word layer; empty root words, duplicate root words, and invalid connectors are deleted by the system during the splitting stage.

[0061] Furthermore, a historical naming reference set is extracted from the historical indicator naming corpus. Using the indicator type, object root, dimension root, and relation root corresponding to the root sequence to be abbreviated as search criteria, the system searches for confirmed standardized English names in the historical indicator naming corpus. When a historical English name has the same indicator type, object root, or dimension root as the root sequence to be abbreviated, that historical English name is included in the historical naming reference set. For duplicate historical English names, the system retains only one; historical English names that have been discontinued or manually marked as non-standard are not included in the historical naming reference set.

[0062] Furthermore, an abbreviation constraint set is extracted from the industry abbreviation database. Each word root in the sequence of words to be abbreviated is read one by one, and the corresponding standard abbreviation word root is queried in the industry abbreviation database. When a word root has a unique standard abbreviation, the standard abbreviation is written into the abbreviation constraint set. When a word root has multiple abbreviation forms, the abbreviation form that is consistent with the applicable business domain and the current indicator and is marked as conflict-free is selected first. When a word root does not have a corresponding abbreviation, the word root is marked as prohibiting the model from creating abbreviations on its own.

[0063] Furthermore, model input prompts are generated based on the sequence of root words to be abbreviated, the historical naming reference set, and the abbreviation constraint set. These prompts include the input name, a list of root words, core root words that must be retained, non-core root words that are allowed to be abbreviated, root words that are prohibited from being replaced, historical naming reference names, standard abbreviation constraints, output format, and the number of candidate names. Core root words include object root words, function root words, and relational root words representing derivation or operational relationships; non-core root words include dimensional root words or conditional root words that can be compressed; and root words that are prohibited from being replaced include those without abbreviation mappings, those with abbreviation conflicts, or those already marked as requiring retention as is. This process confines the model generation task within explicit naming rules.

[0064] Furthermore, the model input prompts are fed into a preset large language model, causing the preset large language model to abbreviate and recombine the initial English names according to the abbreviation constraint set. The model generation follows these rules: core word roots that must be retained are preserved; standard abbreviations from the abbreviation constraint set are used only for non-core word roots that are allowed to be abbreviated; the semantic order of word roots must not be changed; new abbreviations outside the abbreviation constraint set must not be generated; candidate abbreviation names must not be completely identical; each candidate abbreviation name is connected by an underscore. This process ensures that the output of the preset large language model is constrained by the source of word roots, the source of abbreviations, and historical naming patterns. The candidate abbreviation names output by the preset large language model are formatted by removing spaces, repeated underscores, and illegal characters, unifying case to lowercase, and removing completely duplicate candidate abbreviation names. When the number of output abbreviations is less than the preset number of candidate names, word roots that are allowed to be abbreviated are replaced and combined according to the abbreviation constraint set to make up the number of candidate abbreviation names.

[0065] In a specific example, the initial English name generated by S3 is: last 30d paid order chnlorder amt sum. After this initial English name is determined to need abbreviation, the system first splits it into a sequence of root words to be abbreviated: [last 30d, paid order, chnl, order amt, sum]. It then extracts historical English names with the same object root word "order amt", the same function root word "sum", or the same dimension root word "chnl" from the historical indicator naming corpus, forming a historical naming reference set, including: order amt sum, chnl order amt sum, paid order amt. Simultaneously, it extracts an abbreviation constraint set from the industry abbreviation database, where last 30d can be abbreviated to l30d, paid order can be abbreviated to paidord, channel is abbreviated to chnl, amount is abbreviated to amt, and sum remains unchanged. order amt and sum are determined to be core root words that must be retained, last 30d and paid order are non-core root words that can be abbreviated, and chnl is a confirmed abbreviation root word that remains unchanged. Therefore, the model input prompts include: the input name is last 30dpaid order chnl order amt sum; the roots "order amt" and "sum" must be retained; abbreviations of "last30d" → "l30d" and "paid order" → "paid ord" are allowed; the root "chnl" is prohibited; historical reference names are "order amtsum", "chnl order amt sum", and "paid order amt"; the number of candidate names is 3. Under these constraints, the pre-defined large language model outputs the following candidate abbreviations: l30d paid ord chnl order amt sum; l30d chnl paid orderamt sum; paid ord chnl order amt sum. All these candidate abbreviations retain the two core roots "order amt" and "sum," and compress abbreviationable roots based on an industry abbreviation database, while maintaining a naming structure similar to the historical naming reference set. The system then selects a recommended abbreviation from the multiple candidate abbreviations based on a consistency score.

[0066] In S4, recommended abbreviations are selected as the English names to be verified based on the consistency score. Specifically, these include: Calculate the root word retention score, historical pattern matching score, and abbreviation conflict deduction value for each candidate abbreviation name; A consistency score is generated based on the root word retention score, the historical pattern matching score, and the abbreviation conflict deduction value. The candidate abbreviation with the highest consistency score is selected as the recommended abbreviation.

[0067] Specifically, the word root retention score can be calculated using the following expression: ,in, Indicates the first The root word retention score for each candidate abbreviation name; Indicates the number of valid word roots in the initial English name; Indicates the first The retention weights for each word root are as follows: 0.30 for object word roots, 0.25 for function word roots, 0.25 for relation word roots, 0.15 for dimension word roots, and 0.05 for condition word roots. Indicates the first The candidate abbreviation name is for the first... The retention status of each root word is set to 1 when the candidate abbreviation name fully retains the root word, 0.8 when the candidate abbreviation name uses the corresponding standard abbreviation root word in the industry abbreviation dictionary, and 0 when neither the root word is retained nor the standard abbreviation word is used.

[0068] Historical pattern matching scores can be calculated using the following expression: ,in, Indicates the first Historical pattern matching scores for each candidate abbreviation name; This represents a set of historical naming references; This refers to any historical English name in the historical naming reference set; Indicates the first Candidate abbreviations and their historical English names The number of word roots that match in the same order between them; Indicates the first The number of word roots in each candidate abbreviation name; Indicates the historical English name The number of word roots in the text.

[0069] The abbreviation conflict deduction can be calculated using the following expression: ,in, Indicates the first Abbreviation conflict deduction value for each candidate abbreviation name; This indicates a name duplication conflict flag. It is set to 1 when the candidate abbreviation name is exactly the same as the existing standardized indicator name and the corresponding structured indicator elements are inconsistent; otherwise, it is set to 0. This indicates a conflicting abbreviation term. It is set to 1 if any abbreviation root in the candidate abbreviation name corresponds to multiple complete English roots, and 0 otherwise. This indicates a semantically similar conflict marker. When the semantic similarity between the candidate abbreviation name and the existing standardized indicator name reaches 0.90 but the structured indicator elements are inconsistent, it is set to 1; otherwise, it is set to 0. These represent the deduction weights for name duplication conflicts, abbreviation ambiguity conflicts, and semantic similarity conflicts, respectively, with values ​​of 1, 2, 3, and 4 respectively. This expression allows for the quantification and deduction of different types of abbreviation conflicts, making it difficult for candidate abbreviations with naming ambiguities to be selected as recommended abbreviations.

[0070] A consistency score is generated based on root word retention score, historical pattern matching score, and abbreviation conflict deduction. The consistency score can be calculated using the following expression: ,in, Indicates the first Consistency score of each candidate abbreviation name; Indicates the first The root word retention score for each candidate abbreviation name; Indicates the first Historical pattern matching scores for each candidate abbreviation name; Indicates the first Abbreviation conflict deduction value for each candidate abbreviation name; These represent the scoring weights for word root retention score, historical pattern matching score, and abbreviation conflict deduction value, respectively, with the following values: , , This expression establishes a unified scoring standard that balances preserving core semantics, continuing historical naming patterns, and avoiding abbreviation conflicts. After completing the consistency scoring, the system sorts the candidate abbreviations from highest to lowest consistency score and identifies the candidate abbreviation with the highest consistency score as the recommended abbreviation. When two candidate abbreviations have the same consistency score, the system prioritizes the candidate abbreviation with the higher root word preservation score; when the root word preservation scores are still the same, the system prioritizes the candidate abbreviation with the shorter character count; when the character count is still the same, the system prioritizes the candidate abbreviation with the root word order consistent with the historical naming reference set. This process ensures the determinism of the recommended abbreviation selection process and avoids output instability issues when multiple candidate names are listed concurrently.

[0071] S5: Perform semantic deduplication and root consistency verification on the English name to be verified. If it passes, the output is the final standardized indicator name, and the newly generated root mapping relationship is updated to the hierarchical root library. S5 specifically includes: S51: Perform semantic similarity matching between the English name to be verified and the stored standardized indicator names; When the semantic similarity reaches the preset deduplication threshold, the corresponding stored standardized indicator name is output. When the semantic similarity does not reach the preset deduplication threshold, the mapping relationship between the word roots in the English name to be verified and the structured index elements is verified by word root consistency. S52: When the root word consistency check passes, output the English name to be checked as the final standardized indicator name; S53: Write the root mapping relationships of the English names to be verified that are not recorded in the hierarchical root database into the corresponding root layer.

[0072] The aforementioned preset deduplication threshold is specifically set to 0.92. This threshold is obtained through statistical analysis of deduplicated samples of historical standardized indicator names and corresponding indicator processing definitions. Manually confirmed duplicate indicator pairs from the historical indicator database are used as positive samples, while indicator pairs with similar names but different definitions are used as negative samples. The semantic similarity of each indicator pair is calculated. When the threshold is below 0.92, some indicators with similar names but different dimensional constraints will be misjudged as duplicates. When the threshold is above 0.92, some duplicate indicators with only abbreviation differences or differences in word root order cannot be identified. Therefore, setting 0.92 as the preset deduplication threshold achieves a stable boundary between duplicate identification and erroneous merging control. By setting this threshold, the generation of duplicate standardized indicator names can be reduced, while avoiding the erroneous merging of different indicator definitions.

[0073] During deduplication, if the semantic similarity between the English name to be verified and any stored standardized indicator name reaches 0.92, the system determines that they correspond to the same indicator semantics and outputs the stored standardized indicator name, without generating a new final standardized indicator name. If the semantic similarity between the English name to be verified and all stored standardized indicator names is less than 0.92, the system determines that there is no duplicate naming and continues to perform root consistency verification. This process allows deduplication to form clear branches, ensuring that duplicate indicators reuse existing standard names, while non-duplicate indicators enter subsequent verification processes. Furthermore, to prevent misjudgment due to a single semantic similarity, a key field consistency check is performed after reaching a preset deduplication threshold. The key field consistency check includes at least the consistency of the measurement object, the consistency of the aggregation function, and the consistency of the indicator type. When all three are consistent, the deduplication result takes effect; when any one of them is inconsistent, the system does not perform deduplication output and transfers the English name to be verified to root consistency verification. This check process avoids the incorrect merging of indicators with similar root names but different statistical objects or methods, improving the accuracy of the final standardized indicator name.

[0074] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0075] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for automatic decomposition and naming of an indicator system based on semantic parsing, characterized in that, Includes the following steps: S1: Obtain the text of the indicator processing criteria, extract the measurement object, aggregation function, dimension limitation and the identifier of the referenced indicator based on semantic parsing, and generate structured indicator elements; S2: Based on the structured indicator elements, identify the dependencies between the measurement object, aggregation function, dimension constraints, and the referenced indicator identifier, classify the indicators to be named into basic indicators, combined indicators, or derived indicators, and establish indicator reference links. S3: Based on a pre-built hierarchical root word library, structured indicator elements are mapped to root word sequences, and combined with the indicator type and indicator reference link of the indicator to be named, a preliminary English name is generated by splicing them in a preset order. S4: Determine whether the initial English name needs to be abbreviated based on the preset length threshold; If an abbreviation is required, the initial English name, historical indicator naming corpus, and industry abbreviation dictionary are input into the preset large language model to generate candidate abbreviation names, and the recommended abbreviation name is selected as the English name to be verified based on the consistency score; if no abbreviation is required, the initial English name is used as the English name to be verified. S5: Perform semantic deduplication and root consistency verification on the English name to be verified. If it passes, output the final standardized index name and update the newly generated root mapping relationship to the hierarchical root library.

2. The method for automatic decomposition and naming of an index system based on semantic parsing according to claim 1, characterized in that, S1 specifically includes: S11: Perform word segmentation on the text of the indicator processing criteria to obtain a text word sequence; S12: Perform entity recognition and syntactic dependency analysis on the text word sequence, identify words representing statistical objects as measurement objects, and identify words representing statistical methods as aggregation functions; S13: Identify terms representing statistical range, statistical time, statistical conditions, and grouping fields as dimension constraints; S14: Identify existing indicator names, indicator codes, or indicator aliases appearing in the indicator processing caliber text as the identifiers of the referenced indicators; S15: Generate structured indicator elements based on the measurement object, aggregation function, dimension limitation, and referenced indicator identifier.

3. The method for automatic decomposition and naming of an index system based on semantic parsing according to claim 1, characterized in that, The classification of the indicators to be named in S2 into basic indicators, combined indicators, or derived indicators specifically includes: When the structured indicator elements do not contain the identifier of the referenced indicator, but contain the measurement object and the aggregation function, the indicator to be named will be classified as a basic indicator. When a structured indicator element contains at least two referenced indicator identifiers, and there is a computational relationship between the referenced indicator identifiers, the indicator to be named will be classified as a combined indicator. When a structured indicator element contains the identifier of a referenced indicator and includes time offset, ratio conversion, difference calculation, or month-on-month calculation relationship, the indicator to be named will be classified as a derived indicator.

4. The method for automatic decomposition and naming of an index system based on semantic parsing according to claim 3, characterized in that, The establishment of the indicator reference link in S2 specifically includes: Based on the operational, derivational, and dimensional inheritance relationships between the referenced metric identifier and the unnamed metric, a metric reference chain is generated. The metric reference chain includes the current metric identifier, the referenced metric identifier, the reference relationship type, and the reference order. The reference relationship type includes operational reference, derivational reference, and dimensional inheritance reference.

5. The method for automatic decomposition and naming of an index system based on semantic parsing according to claim 1, characterized in that, The hierarchical root word library includes an object root word layer, a function root word layer, a dimension root word layer, a relation root word layer, and an abbreviation root word layer.

6. The method for automatic decomposition and naming of an index system based on semantic parsing according to claim 5, characterized in that, S3 specifically includes: S31: Based on the hierarchical root word library, the measurement objects in the structured indicator elements are mapped to object roots, the aggregation functions are mapped to function roots, and the dimension constraints are mapped to dimension roots; S32: Extract the corresponding type root from the relation root layer according to the indicator type of the indicator to be named; S33: When the indicator to be named has an indicator reference link, extract the corresponding relation root from the relation root layer according to the reference relationship type; S34: Concatenate the root words in the order of dimension, object, function, type, and relation to generate a preliminary English name.

7. The method for automatic decomposition and naming of an index system based on semantic parsing according to claim 1, characterized in that, The preset length threshold in S4 includes a character length threshold of 30 and a word root number threshold of 5. If the number of characters in the preliminary English name is greater than 30, or the number of word roots in the preliminary English name is greater than 5, the preliminary English name is determined to need to be abbreviated. If the number of characters in the preliminary English name is no more than 30 and the number of word roots in the preliminary English name is no more than 5, it is determined that the preliminary English name does not need to be abbreviated.

8. The method for automatic decomposition and naming of an index system based on semantic parsing according to claim 7, characterized in that, The step of inputting the preliminary English name, historical indicator naming corpus, and industry abbreviation dictionary into a preset large language model to generate candidate abbreviation names specifically includes: S41: Perform root word splitting on the preliminary English name to obtain the root word sequence to be abbreviated, and extract historical English names with the same indicator type or the same dimension root words as the root word sequence to be abbreviated from the historical indicator naming corpus to form a historical naming reference set. S42: Extract standard abbreviation roots from the industry abbreviation database that correspond to each root in the sequence of root words to be abbreviated, and form an abbreviation constraint set; S43: Generate model input prompt information based on the sequence of word roots to be abbreviated, the historical naming reference set, and the abbreviation constraint set. The model input prompt information includes core word roots that must be retained, non-core word roots that are allowed to be abbreviated, word roots that are prohibited from being replaced, and the number of candidate names. S44: Input the model input prompt information into the preset large language model, so that the preset large language model performs abbreviation and recombination of the initial English name according to the abbreviation constraint set, and outputs multiple candidate abbreviation names.

9. The method for automatic decomposition and naming of an index system based on semantic parsing according to claim 8, characterized in that, In step S4, the selection of recommended abbreviations as the English names to be verified based on consistency scores specifically includes: Calculate the root word retention score, historical pattern matching score, and abbreviation conflict deduction value for each candidate abbreviation name; A consistency score is generated based on the root word retention score, the historical pattern matching score, and the abbreviation conflict deduction value. The candidate abbreviation with the highest consistency score is selected as the recommended abbreviation.

10. The method for automatic decomposition and naming of an index system based on semantic parsing according to claim 1, characterized in that, S5 specifically includes: S51: Perform semantic similarity matching between the English name to be verified and the stored standardized indicator names; When the semantic similarity reaches the preset deduplication threshold, the corresponding stored standardized indicator name is output. When the semantic similarity does not reach the preset deduplication threshold, the mapping relationship between the word roots in the English name to be verified and the structured index elements is verified by word root consistency. S52: When the root word consistency check passes, output the English name to be checked as the final standardized indicator name; S53: Write the root mapping relationships of the English names to be verified that are not recorded in the hierarchical root database into the corresponding root layer.

Images