Text similarity-based index repeated construction identification method and apparatus

Abstract

The invention provides an index repeated construction recognition method and device based on text similarity, and the method comprises the steps: building a specification standard of index construction repeated recognition according to a stock index in an index library; obtaining element data of the newly added indexes; according to a similarity calculation rule of different elements, obtaining a similarity result between the newly added index and the stock index on different elements; according to a similarity result, obtaining a comprehensive aspect similarity between the newly added index and each stock index; obtaining similarity ranks TOPN of the basic information elements, the description information elements, the logic structure elements and the comprehensive aspect similarity; when the similarity ranking TOPN has the indexes of which the similarity ranking TOPN is greater than a first preset threshold value, determining that the newly added indexes are repeated indexes; and when each similarity ranking TOP1 in the similarity ranking TOPN is smaller than a second preset threshold value, determining that the newly-added index is a non-repeated index, so that the reliability and accuracy of index construction can be improved.

Description

Technical Field

[0001] This article belongs to the field of computer technology, and specifically relates to a method and device for identifying duplicate construction of indicators based on text similarity. Background Art

[0002] In the process of rapid development, enterprises usually cross over into multiple business areas, and as a result, they establish numerous business systems and business processes. These business systems and business processes are usually interdependent, making the management of indicators increasingly complex. First, enterprises may use the same indicator names in different business lines, but the definitions and logic of these indicators may be different, which causes confusion and misunderstanding in data acquisition and increases the difficulty of data analysis. Second, different business departments may develop the same or similar indicators separately, which leads to the problem of duplicate indicator construction, resulting in resource waste and data inconsistency. Third, when enterprises derive other indicators from the original indicators, due to the lack of documentation of indicator development logic, it is difficult for developers to understand or adjust the indicator calculation method when developing new indicators or explaining the original indicators. Therefore, enterprises need to develop indicator similarity recognition technology to cope with the challenges brought by inconsistent indicators.

[0003] Eliminating the duplication of indicators in the communications field has important value: first, it reduces the workload of various departments, allowing the team to focus more on the definition and tracking of core business indicators, which helps to better achieve goals. Second, management will obtain clearer and more consistent data, thereby improving the reliability of decision-making. Most importantly, this move will make the group company's risk prevention and control system more simplified and transparent, reduce the complexity of management and monitoring, and help to more effectively deal with potential risks, thereby promoting the company's high-quality development.

[0004] The existing methods for identifying indicator similarity are mainly based on business experience, which relies on employees' professional knowledge and experience. Employees compare the indicator data of different companies to determine the similarity between them. This method can identify similar indicators to a certain extent, but it also has some obvious shortcomings. For example, employees' professional knowledge is different: different employees will make inconsistent judgments and decisions when comparing the same indicator group; on the other hand, the existing methods require enterprises to invest a lot of human resources, which further increases the human resource costs of enterprises. The existing methods for judging the similarity of indicators are based on the properties of polynomial rings or the decomposition logic of indicators, converting the original indicators into standard calculation formulas or combinations of their internal basic elements. The existing methods are too dependent on the types of indicators, the similarity calculation methods are single and the evaluation is incomplete, with calculation limitations and implementation complexity, and they do not make full use of the indicator text information, which makes them unable to adapt to the complex and changeable enterprise operation needs, and cannot effectively reduce the duplication of indicators, resulting in serious consequences such as waste of enterprise resources. Summary of the invention

[0005] In view of the above problems in the prior art, the purpose of this paper is to provide a method and device for identifying duplicate indicator construction based on text similarity, which can improve the accuracy of indicator construction.

[0006] In order to solve the above technical problems, the specific technical solutions of this article are as follows:

[0007] On the one hand, this paper provides a method for identifying duplicate construction of indicators based on text similarity, the method comprising:

[0008] According to the existing indicators in the indicator library, establish the specification standard for repeated identification of indicator construction, the specification standard is used to represent the similarity calculation rules of the specified element combination in the application field, and the specified element combination at least includes basic information elements, descriptive information elements and logical structure elements;

[0009] Obtain a newly added indicator, and obtain element data corresponding to a specified element combination of the newly added indicator;

[0010] According to the similarity calculation rules of different elements and the element data corresponding to the specified element combination of the newly added indicator, the similarity results between the newly added indicator and the existing indicator on different elements are obtained;

[0011] According to the similarity results, the factor loading matrix of different elements is obtained through principal component analysis;

[0012] According to the factor loading matrix, the comprehensive aspect similarity between the newly added indicator and each existing indicator is obtained;

[0013] Obtain the similarity ranking top 10 of the new indicator and the existing indicator in basic information elements, description information elements, logical structure elements, and comprehensive similarity top 10, where N≥3;

[0014] When there are indicators in the similarity ranking TOPN whose basic information elements, description information elements, logical structure elements and comprehensive aspects are all greater than the first preset threshold, the newly added indicator is determined to be a duplicate indicator;

[0015] When each similarity ranking TOP1 in all the similarity ranking TOPN is less than a second preset threshold, the newly added indicator is determined to be a non-repetitive indicator and is added to the indicator library, wherein the second preset threshold is less than the first preset threshold.

[0016] Furthermore, based on the existing indicators in the indicator database, standard standards for duplicate identification of indicator construction are established, including:

[0017] Determine the application areas corresponding to the indicator library;

[0018] Determine, according to the application field, the indicator-related information in the indicator library, wherein the indicator-related information at least includes indicator definition, calculation logic and generation method;

[0019] According to the indicator-related information, a specified element combination corresponding to the indicator in the indicator library is determined, and a similarity calculation rule for each element in the indicator element combination is established.

[0020] Furthermore, the similarity calculation process for basic information elements is as follows:

[0021] Determine the basic information type corresponding to the basic information element of the indicator in the indicator library;

[0022] According to the basic information type, the newly added indicator and the existing indicator are respectively subjected to word segmentation processing to obtain word segmentation results of the newly added indicator and the existing indicator;

[0023] According to the segmentation results of the newly added indicator and the existing indicator, the similarity results of the newly added indicator and the existing indicator in the basic information elements are obtained by the following formula;

[0024] Among them, S(I_new.types) is the word segmentation result of the basic information elements in the new indicators, S(I_i.types) is the word segmentation result of the basic information elements in the existing indicators, and Sim is the similarity result between the new indicators and the existing indicators in the basic information elements.

[0025] Furthermore, the similarity calculation process for the description information elements is as follows:

[0026] Performing word segmentation processing on the description information elements of the newly added indicator and the existing indicator respectively to obtain word segmentation tag sequences of the newly added indicator and the existing indicator;

[0027] Input the word segmentation token sequences into the ALBERT pre-trained model respectively to obtain an embedding vector sequence corresponding to each word segmentation token sequence;

[0028] Input the embedding vector sequences of the newly added indicators and the existing indicators into the Transformer encoder in the ALBERT model respectively to obtain the encoded embedding vector sequence, and use the first embedding vector in the top-level encoding result in the Transformer encoder as the target embedding vector;

[0029] According to the target embedding vector of the newly added indicator and the target embedding vector of the existing indicator, the similarity between the added indicator and the existing indicator in the description information element is calculated.

[0030] Furthermore, the similarity calculation process for logical structure elements is as follows:

[0031] When the logical structure element is a logical structure text, a standardized element text is extracted from the logical structure text using a text generation technology, wherein the text generation technology at least includes classifying and parsing the logical structure text;

[0032] Performing character conversion, cleaning, and segmentation steps on the standardized element text to obtain target text elements;

[0033] According to the target text elements of the newly added indicator and the existing indicator, the similarity between the newly added indicator and the existing indicator in the logical structure elements is calculated by using the cosine distance calculation method;

[0034] When the logical structure element is a mathematical calculation formula, the variables and mathematical symbols in the mathematical calculation formula are determined;

[0035] For the variables given in the formula, the Word2Vec technology is used to calculate the semantic similarity of the variables between formulas;

[0036] With respect to the mathematical symbols in the formula, by comparing the algebraic expressions, symbol usage and structural composition in the formula, the similarity between the newly added indicators and the existing indicators in terms of logical structural elements is obtained.

[0037] Further, according to the similarity results, the factor loading matrix of different elements is obtained through principal component analysis, which also includes:

[0038] Obtaining the similarity weight of each element of the newly added indicator and each of the existing indicators in the specified element combination;

[0039] According to the similarity weight, and by obtaining the similarity between the newly added indicator and each of the existing indicators in each element of the specified element combination, the final similarity between the newly added indicator and the existing indicators on each element in the specified element combination is obtained.

[0040] Furthermore, the factor loading matrix of different elements is obtained by principal component analysis according to the similarity result, including:

[0041] Obtaining a similarity vector between the newly added indicator and the existing indicators according to the similarity between each element in the specified element combination and the newly added indicator;

[0042] Determine, according to the similarity vector, a representative factor corresponding to each similarity by using a principal component analysis method, wherein the representative factor represents a feature weight of each similarity;

[0043] The factor loading matrix is ​​determined according to the representative factors corresponding to each similarity.

[0044] Furthermore, according to the factor loading matrix, the comprehensive similarity between the newly added indicators and the existing indicators is obtained, including:

[0045] The final similarity between the newly added indicator and the existing indicator on each element in the specified element combination, and the factor loading matrix, obtain the comprehensive aspect similarity between the newly added indicator and the existing indicator.

[0046] Furthermore, the method further comprises:

[0047] When there is at least one similarity among all the similarity rankings TOPN that is less than or equal to the first preset threshold, and there is at least one similarity ranking TOP1 among all the similarity rankings TOPN that is greater than or equal to the second preset threshold, the stock indicator list corresponding to the similarity rankings TOPN of the basic information elements, the stock indicator list corresponding to the similarity rankings TOPN of the descriptive information elements, and the stock indicator list corresponding to the similarity rankings TOPN of the logical structure elements are summarized to obtain a candidate stock indicator list, so as to perform manual judgment based on the candidate stock indicator list.

[0048] On the other hand, this article also provides a device for identifying duplicate construction of indicators based on text similarity, the device comprising:

[0049] A specification standard establishment module is used to establish specification standards for repeated identification of indicator construction based on the stock indicators in the indicator library. The specification standards are used to represent the similarity calculation rules of the specified element combination in the application field. The specified element combination includes at least basic information elements, descriptive information elements and logical structure elements.

[0050] An element data acquisition module is used to acquire a newly added indicator and obtain element data corresponding to a specified element combination of the newly added indicator;

[0051] A similarity calculation module, used to obtain similarity results between the new indicator and the existing indicator on different elements according to similarity calculation rules of different elements and element data corresponding to the specified element combination of the new indicator;

[0052] A factor loading matrix acquisition module is used to obtain the factor loading matrix of different elements through principal component analysis according to the similarity result;

[0053] A comprehensive aspect similarity calculation module, used to obtain the comprehensive aspect similarity between the newly added indicator and each existing indicator according to the factor loading matrix;

[0054] A screening module is used to obtain the similarity ranking top 10 of the new indicator and the existing indicator in basic information elements, description information elements, logical structure elements, and comprehensive similarity top 10, where N≥3;

[0055] The first judgment module is used to determine that the newly added indicator is a duplicate indicator when there are indicators in the similarity ranking TOPN whose basic information elements, description information elements, logical structure elements and comprehensive aspects are all greater than the first preset threshold value;

[0056] The second judgment module is used to determine that the newly added indicator is a non-repetitive indicator and add it to the indicator library when each similarity ranking TOP1 in all the similarity ranking TOPN is less than a second preset threshold, wherein the second preset threshold is less than the first preset threshold.

[0057] Using the above technical solution, a method and device for identifying duplicate indicator construction based on text similarity is described in this article. The method includes: establishing a standard for duplicate indicator construction identification based on the existing indicators in the indicator library, the standard is used to represent the similarity calculation rules of the specified element combination in the application field, and the specified element combination at least includes basic information elements, descriptive information elements and logical structure elements; obtaining a new indicator and obtaining the element data corresponding to the specified element combination of the new indicator; according to the similarity calculation rules of different elements and the element data corresponding to the specified element combination of the new indicator, obtaining the similarity results between the new indicator and the existing indicators on different elements; according to the similarity results, obtaining the factor loading matrix of different elements through principal component analysis; according to the factor loading matrix, obtaining the comprehensive similarity between the new indicator and each existing indicator; obtaining the The newly added indicator and the existing indicator are ranked in the TOPN similarity of basic information elements, TOPN similarity of descriptive information elements, TOPN similarity of logical structure elements, and TOPN similarity of comprehensive aspects, where N≥3; when there are indicators in the similarity ranking TOPN whose similarities in basic information elements, descriptive information elements, logical structure elements, and comprehensive aspects are all greater than a first preset threshold, the newly added indicator is determined to be a duplicate indicator; when each similarity ranking TOP1 in all the similarity ranking TOPN is less than a second preset threshold, the newly added indicator is determined to be a non-duplicate indicator and is added to the indicator library, where the second preset threshold is less than the first preset threshold. The method provided in this article simplifies the indicator disassembly steps by dividing and analyzing the elements of each indicator, reduces the complexity and uncertainty of the disassembly process, and helps to ensure the consistency and comparability of the indicators. At the same time, this technical solution adopts different similarity calculation technologies according to the characteristics of the element text, rather than relying on just one calculation method, and obtains comprehensive aspect similarity based on principal component analysis (PCA), which enables it to more comprehensively evaluate the similarity of indicators and more accurately capture the differences between indicators, solving the problem of inaccurate recognition caused by the small number of indicator similarity evaluation dimensions and insufficient use of semantic information in the existing technology.

[0058] In order to make the above and other purposes, features and advantages of this article more obvious and easy to understand, the following specifically cites preferred embodiments and describes them in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of this article or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of this article. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0060] Figure 1 A schematic diagram of the steps of a method for identifying duplicate construction of indicators based on text similarity provided in an embodiment of this invention is shown;

[0061] Figure 2 A schematic diagram of the framework of the method provided in the embodiment of this article is shown;

[0062] Figure 3 A schematic diagram of the working logic of the ALBERT pre-training model in an example of this article is shown;

[0063] Figure 4 A flowchart of similarity calculation of logical structure elements in an example of this article is shown;

[0064] Figure 5 A flowchart of comprehensive aspect similarity calculation in an example of this article is shown;

[0065] Figure 6 A schematic diagram of the indicator duplication construction process in an example of this article is shown;

[0066] Figure 7 A schematic diagram of the structure of a computer device provided in an embodiment of this invention is shown. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of this article to clearly and completely describe the technical solutions in the embodiments of this article. Obviously, the described embodiments are only part of the embodiments of this article, not all of the embodiments. Based on the embodiments of this article, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this article.

[0068] It should be noted that the terms "first", "second", etc. in the specification and claims of this article and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of this article described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0069] First, let’s explain the technical terms that appear below:

[0070] Metrics: Metrics are standardized measurements used to measure, evaluate, or describe a phenomenon, property, or effect. In data analysis, research, and evaluation, metrics are often used to measure a specific property, performance, or result, such as error rate, accuracy, efficiency, etc.

[0071] Duplicate construction of indicators: Duplicate construction of indicators refers to the creation or definition of similar or identical indicators multiple times in enterprise data analysis and indicator management, resulting in a waste of enterprise resources.

[0072] Text similarity: Text similarity is a method to measure the similarity between text documents. It is used to determine the semantic or structural similarity between texts and is usually expressed as a value between 0 and 1, where 0 means completely dissimilar and 1 means exactly the same.

[0073] Standardized element text: Standardization refers to formatting or processing indicator data according to certain rules or standards to ensure consistency and comparability. Element text is the basic building block in the indicator specification, usually representing a description of a specific aspect of the indicator, such as indicator code, indicator name, indicator explanation, indicator classification and other information.

[0074] Indicator standardization: newly added indicators will be organized according to indicator specifications to ensure that their element texts meet the standards of their fields.

[0075] Basic information elements: Basic information elements refer to the core and basic information related to the indicator, such as indicator code, indicator name, indicator type, etc. This information is usually concise and appears in a shorter text form.

[0076] Descriptive information elements: detailed text information involving the meaning, explanation, expected effects and usage scenarios of the indicator. These contents are used to explain and elaborate the meaning and use of the indicator, and usually appear in the form of longer text.

[0077] Logical structural elements: structural texts involving mathematical expressions, programming languages, specific business logic description languages, etc. These elements contain domain-specific symbols, hierarchical structures, and grammatical rules, and are suitable for similarity calculation of derived indicators or composite indicators.

[0078] PCA (Principal Component Analysis): PCA is a multivariate statistical analysis method used to reduce data dimensionality and reveal patterns in data. It looks for principal components in the data, which are linear combinations of the original data features in order to reduce the complexity and noise of the data.

[0079] ALBERT (A Lite BERT, ALBERT pre-trained model): is a pre-trained model for natural language processing (NLP) tasks, which is an improved version of BERT (Bidirectional Encoder Representations from Transformers); pre-trained models are neural network models trained through self-supervised learning on large-scale text data. They learn the language knowledge and context in the text, and can then be used for fine-tuning of various NLP tasks.

[0080] Embedding vector: An embedding vector is a mathematical representation that maps text data (such as words, sentences, or characters) to a continuous vector space. In BERT, the embedding vector is a representation at the word or subword level so that the model can better understand the meaning of the language. These embedding vectors are learned through a pre-trained neural network model (usually a Transformer architecture), where each word or subword corresponds to a high-dimensional vector that captures grammatical and semantic information.

[0081] Grammar generation technology: Grammar generation technology is a method for automatically generating text, code, structure or data according to specific grammatical rules or patterns. These rules are usually based on formal grammars, templates or patterns to extract or generate useful information from given input.

[0082] Jaccard Similarity: Jaccard similarity is an indicator used to measure the similarity between two sets, usually expressed as J(A,B). It is calculated by dividing the number of elements in the intersection of the two sets by the number of elements in their union, that is, J(A,B) = |A∩B| / |A∪B|. The value range of Jaccard similarity is between 0 and 1, where 0 means that the two sets are completely dissimilar and 1 means that the two sets are exactly the same.

[0083] The existing methods for judging the similarity of indicators are based on the properties of polynomial rings or the decomposition logic of indicators, converting the original indicators into standard calculation formulas or combinations of their internal basic elements. The existing methods are too dependent on the types of indicators, the similarity calculation methods are single and the evaluation is incomplete, with calculation limitations and implementation complexity. They do not make full use of the indicator text information, resulting in their inability to adapt to the complex and changing business operation needs, and they cannot effectively reduce the duplication of indicators, resulting in serious consequences such as waste of enterprise resources.

[0084] In order to solve the above problems, the embodiments of this article provide a method for identifying duplicate indicator construction based on text similarity, which can improve the efficiency of indicator construction. Figure 1This is a schematic diagram of the steps of a method provided in the embodiments of this article. This specification provides the method operation steps described in the embodiments or flowcharts, but based on routine or non-creative work, more or fewer operation steps may be included. The order of steps listed in the embodiments is only one way of executing the steps among many orders, and does not represent the only order of execution. When the actual system or device product is executed, it can be executed in the order of the method shown in the embodiments or the drawings or in parallel. Specifically, Figure 1 As shown, the method may include:

[0085] S101: Based on the stock indicators in the indicator library, establish a specification standard for repeated identification of indicator construction, wherein the specification standard is used to represent a similarity calculation rule of a specified element combination in an application field, wherein the specified element combination includes at least a basic information element, a description information element, and a logical structure element;

[0086] S102: Acquire a newly added indicator, and obtain element data corresponding to a specified element combination of the newly added indicator;

[0087] S103: Obtain similarity results between the new indicator and the existing indicator on different elements according to similarity calculation rules of different elements and element data corresponding to the specified element combination of the new indicator;

[0088] S104: Obtaining factor loading matrices of different elements through principal component analysis according to the similarity results;

[0089] S105: Obtaining the comprehensive aspect similarity between the newly added indicator and each existing indicator according to the factor loading matrix;

[0090] S106: Obtain the similarity ranking top 10 of the new indicator and the existing indicator in basic information elements, description information elements, logical structure elements, and comprehensive similarity top 10, where N≥3;

[0091] S107: When there are indicators in the similarity ranking TOPN whose basic information elements, description information elements, logical structure elements and comprehensive aspects are all greater than the first preset threshold, the newly added indicator is determined to be a duplicate indicator;

[0092] S108: When each similarity ranking TOP1 in all the similarity ranking TOPN is less than a second preset threshold, the newly added indicator is determined to be a non-repetitive indicator and added to the indicator library, wherein the second preset threshold is less than the first preset threshold.

[0093] It can be understood that: the embodiments of this specification establish a set of indicator specification systems, which simplify the steps of disassembling indicators by dividing and analyzing the elements of each indicator, reduce the complexity and uncertainty of the disassembly process, and help ensure the consistency and comparability of indicators. At the same time, the technical solution adopts different similarity calculation technologies according to the characteristics of the element text, rather than relying on only one calculation method, and obtains comprehensive aspect similarity based on principal component analysis (PCA), which makes it more comprehensive. The similarity of indicators can be more accurately captured, and the problem of inaccurate recognition caused by the small number of indicator similarity evaluation dimensions and the insufficient use of semantic information in the prior art is solved. It solves the problems existing in the existing indicator similarity judgment method, improves the accuracy and adaptability of indicator similarity judgment, and thus reduces the waste of enterprise resources.

[0094] Furthermore, based on the existing indicators in the indicator database, standard standards for duplicate identification of indicator construction are established, including:

[0095] Determine the application areas corresponding to the indicator library;

[0096] Determine, according to the application field, the indicator-related information in the indicator library, wherein the indicator-related information at least includes indicator definition, calculation logic and generation method;

[0097] According to the indicator-related information, a specified element combination corresponding to the indicator in the indicator library is determined, and a similarity calculation rule for each element in the indicator element combination is established.

[0098] Among them, the application field can be a specific business field involved in an enterprise or an indicator library, etc. The indicator element similarity calculation module implemented in the embodiment of this specification provides a similarity calculation technology for different element text characteristics. The method may include: for basic information elements, split the element text into a word set, and use the Jaccard formula to calculate the similarity; for descriptive information elements, input the element text into ALBERT, and calculate the similarity based on the text representation vector generated by the model; for logical structure elements, use grammar generation technology to extract structured data in the element text, and calculate the similarity based on the structured data and corresponding domain rules.

[0099] The present specification also provides a method for identifying duplicate construction of indicators based on text similarity, such as Figure 2 As shown, including:

[0100] Step S11: Indicator definition standardization module: First, gather and summarize the experience of experts in various fields, and obtain the indicator definition, calculation logic, generation method and other information related to each field. According to the collected expert experience, organize and summarize, establish a set of indicator specification system applicable to multiple fields, wherein the indicator specification system defines each indicator based on three core elements to improve the consistency of indicator management. In this embodiment, the three core elements include: "basic information element", "descriptive information element" and "logical structure element". "Basic information element" refers to the core and basic information related to the indicator, which is usually concise and appears in a shorter text form. In this embodiment, it includes: indicator code, indicator name, indicator classification, indicator type, indicator attribution unit, indicator proposer, indicator effective time, indicator expiration time, indicator cycle, indicator unit, etc. "Descriptive information element" involves detailed text information such as the meaning, explanation, expected effect and usage scenario of the indicator, which is used to explain and explain the meaning and use of the indicator, and usually appears in a longer text form. In this embodiment, it includes: indicator business measurement caliber, indicator explanation, indicator significance, indicator expected effect, etc. "Logical structural elements" involve the internal structure and composition of indicators, usually containing domain-specific symbols, hierarchical structures, and grammatical rules, and are suitable for similarity calculation of derived indicators or composite indicators. In this embodiment, it includes: the calculation formula of the indicator, the SQL logic for indicator generation, the association conditions for indicator generation, the data collection method of the indicator, etc.

[0101] For each group of new indicators and existing indicators, the element text under its core elements can be obtained according to the indicator specification system. The element text is obtained through standardization: In this embodiment, the standardization processing steps include: naming the indicator name and indicator code according to the unified specification, standardizing the indicator type and calculation logic, and standardizing other indicator element texts.

[0102] In this embodiment, the specific process of obtaining the element text of the newly added indicator and the existing indicator includes:

[0103] Obtain indicator data to be compared by similarity, wherein the indicator data to be compared by similarity includes but is not limited to definition data of the indicator, calculation formula of the indicator, source information of the indicator, attribution and management information of the indicator, etc.

[0104] By performing data cleaning on the acquired indicator data, invalid and erroneous indicator data are removed to obtain a cleaned indicator data set. Correspondingly, the data cleaning method includes the following: by comparing multiple reliable data sources, using expert experience correction or using automated tools to identify and correct erroneous data; detecting duplicate data through automated tools to ensure that the indicator data is unique; through standardized data formats, ensuring that the indicator effective time, indicator update frequency, and indicator evaluation time are unified as ISO standards; naming the indicator names according to unified specifications to eliminate name differences for the same indicator caused by different naming habits.

[0105] As a preference, unified and standardized naming includes the following steps: specifying unified naming rules, which should include the naming structure, the characters and formats allowed for naming, and the case distinction of naming; constructing an indicator name structure, where the indicator name consists of multiple levels of information to express the specific dimensions or classifications of the indicator. Each level of information should have clear semantics, and avoid using abbreviations unless the abbreviations are widely recognized and easy to understand; establishing and maintaining a standardized glossary based on the experience of experts in various fields, and all indicator names must be derived from this glossary. When the company's business development or market changes lead to the addition or adjustment of terms, the glossary should be updated in a timely manner and the affected indicator names should be revised.

[0106] Based on the purpose, use and characteristics of indicator construction, construct the indicator classification standard for a specific field. The indicator classification standard lists the parent category of the indicator and its subordinate subcategories in detail to accommodate all possible indicators. Classify the cleaned indicator data set to ensure that the indicator can be accurately classified after being defined by the classification standard.

[0107] Based on the industry standards in the field and the company's known business rules, define clear and unified calculation formulas for each type of indicator to ensure that the calculation formulas can be uniformly applied throughout the organization in the field and that these formulas can be strictly enforced, including ensuring the mathematical methods used in the calculations, the logical structures used in the calculations, and the tools used in the calculations. The parameters in the calculation formulas can be clearly identified by the parameter manual created by experts, which includes the parameter name, parameter symbol, parameter source, parameter valid range, and parameter specific meaning.

[0108] After the indicators are organized according to the indicator specification system, the element texts have basically met the standards of the corresponding fields, and step S12 is executed to analyze the characteristics of the indicator element texts.

[0109] Step S12: Preferably, by analyzing the indicator elements, the indicator elements are divided into basic information elements, descriptive information elements, and logical structure elements. Figure 2In the similarity calculation process of step S12, different similarity calculation methods are used according to the characteristics of the element text to obtain the similarity results of the newly added indicators and the existing indicators on each element.

[0110] Among them, different similarity calculation methods are used according to the characteristics of the element text to obtain the similarity results, including:

[0111] Step S121: Descriptive information element similarity calculation submodule: Descriptive information elements mainly include detailed text information such as indicator explanation, indicator significance, and indicator usage scenarios. Such elements usually contain multiple words and sentences, and contain rich contextual information and semantic meanings, which makes the evaluation of indicator similarity complicated. Considering that the pre-trained model can capture the semantic information and contextual relationship of the text, the embedding vector generated by the pre-trained model can be used to realize the semantic similarity comparison of the descriptive information elements, and the extracted semantic representation vector contains the deep semantic meaning behind the indicator. The method further divides step S121 into step S1211, step S1212, step S1213, and step S1214.

[0112] Step S1211: For details, please refer to Figure 3 ,First, obtain the descriptive information elements of the ALBERT pre-trained model and ,each set of indicators.

[0113] Step S1212: Segment the description information element text into characters, and convert the description information element text containing several characters into vector representation.

[0114] In this embodiment, the indicator "operating income completion rate" is used to illustrate the process of constructing the semantic representation vector. The indicator explanation belongs to the descriptive information element. The indicator explanation of "operating income completion rate" is: "Operating income: business income refers to the after-tax income reflected in the financial accounts, excluding income from renting houses and selling goods, and excluding income from internal transactions and sales." After the element text is segmented by the word segmenter, it becomes the sequence of "[CLS] Operating income: business income refers to the after-tax income reflected in the financial accounts, excluding income from renting houses and selling goods, and excluding income from internal transactions and sales. [SEP]". Furthermore, the sequence can be input into the ALBERT pre-trained model to obtain an embedded vector sequence after each token in the sequence is mapped. For ease of understanding, the embedded vector sequence is E = (T1, T2,…, T50). The above embedded vector sequence includes a total of 50 embedded vectors, and the embedded vector corresponding to the i-th token is denoted as Ti.

[0115] Step S1213: Input the above embedding vector sequence into the Transformer encoder in the ALBERT model, including a multi-layer self-attention mechanism and a feedforward neural network layer. Each layer encodes the embedding vector to capture the contextual information and semantic relationship of the text. For ease of understanding, H(I) is used to represent the embedding vector sequence after encoding the Ith layer, H(0) represents the embedding vector sequence, and H(L) represents the encoding result of the top layer. Layer1 to LayerL represent the encoders of each layer, with a total of L layers. Then H(0) = E, H(1) = Layer1(H(0)), H(2) = Layer2(H(1))...H(L) = LayerL(H(L-1)).

[0116] Step S1214: In the pre-training stage of the ALBERT model, the embedding vector of the [CLS] tag captures the semantic information of the entire text, so the vector output at this position contains the semantic information of the entire element text. The embedding vector of the [CLS] tag corresponding to the indicator explanation element text in each group of indicators is extracted. This vector can be recorded as the semantic vector corresponding to the element text, which can be expressed as CLS=H(L)[0] in a mathematical formula, where H(L)[0] represents the first embedding vector in the top-level encoding result. Finally, the text similarity is calculated based on the embedding vector to obtain the semantic similarity of each group of indicators in this element text.

[0117] Optionally, in order to better adapt to the context of various industry verticals, the pre-trained model can be fine-tuned in a specific field to capture subtle differences in the text describing information elements. Moreover, the pre-trained model is not limited to the ALBERT model. As other feasible solutions, the pre-trained model can be any one of Sentence-BERT, SimBERT, SimCSE, etc. It is a known technology to use Sentence-BERT, SimBERT, SimCSE and other models for fine-tuning in specific fields, which will not be repeated here. In addition, the results output by the pre-trained model are not limited to the embedded vector after text encoding. The results output by the prediction model can be classification results and prediction probabilities for identifying whether texts are similar.

[0118] Step S122: Basic information element similarity calculation submodule: The basic information elements mainly include indicator classification, indicator cycle, indicator type, indicator unit, indicator owner, indicator dimension and other basic information related to the indicator. Such elements usually have a short text length, are usually concise, and have relatively fixed content, and can generally be selected from predefined tags. Considering that such elements have fewer words and that a quick comparison of such elements can narrow the search scope, the Jaccard similarity is used to calculate the similarity of the basic information elements.

[0119] In this embodiment, the basic information element calculation process is explained by taking indicator classification as an example. For each group of indicators' "indicator classification" text, the indicator classification types of the newly added indicator I_new and the existing indicator I_i are first segmented. Text segmentation methods include forward maximum matching method (MM), reverse maximum matching method (RMM), and matching method based on hidden Markov model (HMM). The above text segmentation methods are known algorithms and will not be repeated here. After obtaining the segmentation results of the indicator classification, S(I_new.types) is recorded as the segmentation result of the indicator classification in the newly added indicator, and S(I_i.types) is recorded as the segmentation result of the indicator classification in the existing indicator.

[0120] The text similarity between the new indicators and the existing indicators in the indicator classification is calculated according to the following formula:

[0121]

[0122] The above Sim value is the similarity calculation result of the indicator classification.

[0123] Step S123: Logical structure element similarity calculation submodule: The logical structure element similarity calculation submodule mainly processes text information with strong structure, including expressions of mathematical operations, programming languages, business logic description languages, etc. If the logical structure element involves a computer programming language or a specific business logic language, step S123 is executed, and the method further divides step S123 into step S1231, step S1232, step S1233, and step S1234. If the logical structure element involves an expression of mathematical operations, step S124 is executed, and the method further divides step S124 into step S1241, step S1242, step S1243, and step S1244.

[0124] Step S1231: For details, please refer to Figure 4In this embodiment, the indicator generation method in the field of database design and data warehouse is taken as an example to illustrate the calculation process of the text similarity of logical structure elements. The indicator generation method in this field is based on the SQL syntax in the computer programming language. Such elements include fact table name, fact table field, fact table screening condition, calculation logic, dimension table name, dimension table field, dimension table screening condition, association condition, SQL logic, and indicator calculation configuration. In this embodiment, in order to comprehensively map the actual situation of the business with a multidimensional data model, the indicator generation method can be further decomposed into four standardized parent elements: measurement value technical caliber, dimension value technical caliber, association relationship, and SQL logic. Each type of parent element is divided into different standardized child elements. Before calculating the similarity of such elements, it is necessary to evaluate the integrity of the logical structure elements in step S11, that is, to obtain the missing rate of the logical structure element text. The calculation expression of the missing rate is: missing rate = number of missing elements / total number of elements. .

[0125] Step S1232, by evaluating the integrity of such elements, it can be found how many element texts are missing in the logical structure element text. If such element texts are missing, the grammar generation technology can be used to extract standardized element texts from the SQL logic to fill in the missing elements; if it is found in the integrity check that the standardized element text of the indicator meets the requirements, the similarity of the logical structure elements can be directly calculated based on the standardized element text.

[0126] The steps of extracting standardized element text from SQL logic using grammar generation technology include: an indicator classification step and a code parsing step.

[0127] The indicator classification step includes dividing the enterprise production indicators into atomic indicators, derived indicators and composite indicators. Atomic indicators refer to those metrics based on specific business event behaviors. These metrics are regarded as indicators that cannot be decomposed in the business definition. They are clear nouns with independent business meanings. In this embodiment, "after-tax amount" can be regarded as an atomic indicator; derived indicators are those indicators that have actual business meanings but cannot be directly obtained from the middle layer of the data warehouse. They need to meet business needs through a bridge connecting the middle layer of the data warehouse and the business party. In this embodiment, "step income increase" can be regarded as a derived indicator; composite indicators are obtained by combining or calculating multiple atomic indicators or derived indicators, and are used to comprehensively evaluate or measure the situation of a certain business or performance. In this embodiment, operator DOU (daily active users), operator MOU (monthly active users), operating income completion rate, etc. can all be regarded as composite indicators.

[0128] For the SQL logic of atomic indicators, the parsing method includes: using SQL parser or regular expression to determine the SELECT clause, FROM clause, WHERE clause, JOIN clause and other clauses. By decomposing the data tables and fields of the FROM clause, JOIN clause, WHERE clause, and ON clause, the fact table name, fact table screening conditions and dimension table name, dimension table screening conditions, dimension table fields, and association conditions of the atomic indicator are determined. By decomposing the fields of the SELECT clause, the fact table fields of the atomic indicator are determined;

[0129] For the SQL logic of derived indicators, the parsing methods include: in addition to extracting elements by decomposing clauses as mentioned above, you can also determine the calculation logic in the derived indicators by decomposing the SELECT clause and sorting functions such as rank(), rownumber(), or aggregation functions sum(), max(), min(), etc. The above functions allow you to obtain necessary data from the middle layer of the data warehouse and convert it into derived indicators with actual business meanings.

[0130] Regarding the similarity comparison of composite indicators, since composite indicators can be regarded as obtained by mathematical operations on atomic indicators or derived indicators, the standardized elements can be first extracted based on the above-mentioned grammar generation technology, and then the comparison method of mathematical expressions mentioned in the following step S124 can be used to achieve similarity comparison in a mathematical sense.

[0131] Step S1233: Performing data processing procedures such as English uppercase and lowercase conversion, English text cleaning, word segmentation, etc. on the standardized element text;

[0132] Step S1234: Select the cosine distance calculation formula or the Jaccard formula to calculate the text similarity of each group of indicators with respect to the standardized element text. Specifically, the steps for calculating the similarity of the logical structure elements are as follows: construct a feature vector based on the standardized element text, and each dimension of the feature vector can represent a standardized element. Then the standardized element sets of the newly added indicators and the stock indicators can be respectively represented by A=(a 1 ,a 2 ,a 3 ,...a n ) and B=(b 1 ,b 2 ,b 3 ,...b n ), using binary code to indicate whether each element in the indicator to be tested is consistent, for each sub-element a i and b j Calculate the similarity between them. The similarity calculation formula is cos_sim = (A·B) / (‖A‖·‖B‖), and get the similarity vector C = (c 1 ,c2 ,c 3 ,...c n ), the similarity vector is a binary vector, and the value of each dimension of the vector represents the consistency of the indicator to be tested under this dimension. Finally, the obtained similarity is input into the following formula for normalization processing for comparison and analysis:

[0133] Among them, w ij Is the standardized element a i and standardized element b i The weight of .

[0134] Step S124: In this embodiment, the logical structure element may be related to the calculation formula of the indicator. This type of element text usually follows the calculation rules of mathematical expressions, focusing more on mathematical meaning rather than semantic meaning, which makes this type of element text different from natural language text in syntax and structure. In order to better extract the formula structure features, grammar generation technology is used to help compare the similarity of the calculation formula of each group of indicators. The method further divides step S124 into step S1241, step S1242, step S1243, and step S1244.

[0135] Step S1241: For details, please refer to Figure 4, in step S1241, for the calculation formula of each indicator, the calculation rules described in natural language are converted into mathematical symbolic expressions. Mathematical symbolic expressions include mathematical operators, variables, functions, etc. For example, for the "operating income completion rate" indicator, the operating income completion rate is mainly used to evaluate the comparison between the company's actual income and expected income in a certain period, "operating income completion rate" = ("fixed income completion value") / ("fixed income target value") * 100%, the variables contained in the indicator include "fixed period income completion value" and "fixed period income target value", and the operator in the indicator is division; for the "average user income" indicator, "average user income" = (∑ (monthly income)) / ("number of users in a fixed period"), the operator in the indicator is division, and the function in the indicator is an accumulation function, and the "total income for a fixed period" is the sum of multiple months of "monthly income" input into The indicator "Monthly User Activity" is obtained from the cumulative function; for the "Monthly User Activity" indicator, "Monthly User Activity" = α × (monthly call duration / call duration benchmark) + β × (monthly data usage / data usage benchmark) + γ × (monthly SMS volume / SMS volume benchmark), the variables contained in the indicator include monthly call duration, call duration benchmark, monthly data usage, data usage benchmark, monthly SMS volume, SMS volume benchmark, constants are α, β, γ, the operators in the indicator are multiplication and addition, and the monthly user activity is mainly used to measure the user's activeness in using telecommunications services within a month. It combines the user's call duration, data usage and SMS volume to fully reflect the user's activeness. Step S1242: For the variables in the formula, use Word2Vec technology to calculate the semantic similarity of the variables between formulas. Using Word2Vec technology to calculate semantic similarity is a known technology and will not be repeated here. If the semantic similarity is higher than the preset threshold K, these variables in the formula can be considered similar, and the two similar variables are replaced with a unified mathematical symbol.

[0136] Step S1243: For each mathematical symbolic expression, a simplification operation is performed to simplify the formula and reduce the complexity to obtain a simplified index calculation formula. This step involves the application of algebraic simplification rules or related mathematical principles. The algebraic simplification rules include combining like terms, factoring polynomials, distributive laws of combined terms, etc. Since such mathematical principles belong to known technologies, they will not be elaborated here.

[0137] Step S1244: For the simplified mathematical symbolic expressions, mathematical expressions are compared, including the following three aspects: formula algebra comparison, formula symbol comparison, and formula structure comparison. The formula algebra comparison is: when performing similarity calculations, their mathematical expressions are compared to determine whether they are equivalent or similar in algebraic structure; the formula symbol comparison is: when performing similarity calculations, the symbols used in the formulas are compared to determine whether they are consistent in mathematical symbols; the formula structure comparison is: when performing similarity calculations, the structures of the mathematical expressions are compared to determine whether they are consistent in mathematical structure; if the comparison results of the two groups of calculation formulas in any of the above aspects are dissimilar, then the similarity results of the logical structure elements are dissimilar, otherwise they are similar. The following takes "user monthly activity" (UMA) and "user monthly usage" (UMU) as examples to illustrate the specific process of similarity calculation of logical structure elements in this scenario:

[0138] "User Monthly Activity" (UMA) = α × (monthly number of calls / benchmark monthly number of calls) + β × (monthly data usage / benchmark data usage + monthly SMS sending / benchmark SMS sending)

[0139] "User Monthly Usage" (UMU) = δ × (Current Month Call Time + Base Month Call Time) / Base Month Call Time) + ε × (Current Month Data Traffic - Base Data Traffic) / Base Data Traffic

[0140] The variables extracted from the indicators (UMA) and (UMU) are: number of calls per month, number of calls per base month, monthly data usage, base data usage, number of SMS messages sent per month, base SMS messages sent, talk time per month, base monthly talk time, data traffic per month, base data traffic.

[0141] The word semantic similarity of the above variables is calculated by using Word2Vec technology. Since the similarity of "monthly data usage" and "monthly data flow", "benchmark data usage" and "benchmark data flow" in the variables is higher than the threshold of 0.5, these variables can be replaced by unified mathematical symbols. After the formula is replaced by unified mathematical symbols, the expression becomes:

[0142] “User Monthly Activity” (UMA) = α×(Tm / Tb)+β×(Dm / Db+Sm / Sb)

[0143] “User Monthly Usage” (UMU) = δ×((Am+Ab) / Ab)+ε×((Dm-Db) / Db)

[0144] After simplifying the above expression according to the principle of algebra, the expression becomes:

[0145] “User Monthly Activity” (UMA) = α×(Tm / Tb)+β×(Dm / Db)+β×(Sm / Sb)

[0146] "User Monthly Activity" (UMU) = δ×(Am / Ab)+ε×(Dm / Db)+δ-ε

[0147] The comparison process of mathematical expressions includes: formula algebra comparison, formula symbol comparison, and formula structure comparison.

[0148] The algebraic comparison of formulas mainly determines the consistency of the corresponding terms in (UMA) and (UMU). First, the two expressions are decomposed and the terms in the corresponding positions are extracted, including multiplication terms and addition terms. The terms in the corresponding positions are compared one by one, and the proportion of equal terms in the two expressions is calculated by using the symbolic calculation library. The calculation expression is: the number of equal terms in the expression = the total number of terms in the expression. A threshold is set to determine whether the terms of the two expressions are close enough to be considered similar.

[0149] Formula symbol comparison mainly judges all symbols and operators in (UMA) and (UMU). First, all symbols and operators in the two expressions are extracted, and the symbols and operators in the two expressions are compared one by one, and the number of matches and mismatches is recorded. The Jaccard similarity coefficient can be used to obtain the consistency degree of symbols and operators in the two expressions. The expression is J(U,V)=|U∩V| / |U∪V|, where U and V represent the sets of symbols and operators in the two expressions respectively.

[0150] The formula structure comparison mainly judges the arrangement order and combination of each item of (UMA) and (UMU), and converts the expression into an abstract syntax tree (AST) by using the symbolic calculation library. By traversing this tree, the structural information of the expression can be extracted. In this process, the abstract syntax tree (AST) records the operators and operation trees, and also pays attention to the order and combination relationship in the expression. Using the structural mapping generated by the abstract syntax tree, for expressions with the same structure and different numerical values, their structural mappings should be similar. Using the abstract syntax tree (AST) to calculate the structural similarity of two expressions is a known technology and will not be repeated here.

[0151] Step S13: Multidimensional similarity calculation module: According to the similarity of basic information elements, descriptive information elements, and logical structure elements of each group of indicators, the above similarity sets are weighted averaged respectively, and principal component analysis (PCA) is used to perform multidimensional similarity fusion calculation to obtain the comprehensive similarity score of each group of indicators.

[0152] For details, please refer to Figure 5, obtain the similarity of basic information elements {d1, d2, d3, ..., dn}, the similarity of description information elements {c1, c2, c3, ..., cm}, and the similarity of logical structure elements {j1, j2, j3, ..., jk}, and take the weighted average of the above similarity sets respectively,

[0153] The corresponding expression is w i The weight can be derived based on the professional experience in the corresponding field. The weight is usually between 0 and 1, and the sum of all weights is 1. The corresponding similarity value is input into the expression to obtain the final similarity of the basic information element (sim_basic), the final similarity of the descriptive information element (sim_desc), and the final similarity of the logical structure element (sim_logic).

[0154] Get the similarity combination of all element texts and generate vector P = (d1, d2, d3, ..., jk). Since the similarity value is already in the (0, 1) interval, the similarity value no longer needs to be centered and normalized. Use the principal component analysis (PCA) method to find n representative factors in the P vector that can explain 90% of the data variation. Use F1, ..., Fn to represent each representative factor, where n is the number of principal components.

[0155] Obtain the factor loading matrix, which shows the relationship between the original variables and the principal components, and obtain the feature weights of each element text. The factor loading matrix reflects the correlation between each principal component and the original variable, that is, the feature weight corresponding to each principal component. Finally, the comprehensive similarity of the indicators is obtained based on the feature weights of each element text and the similarity of each element text.

[0156] Specifically, for the i-th original variable and the j-th principal component, its loading ij It can be calculated as follows:

[0157]

[0158] where var_weight ij is the weight of the i-th original variable in the j-th principal component, eigenvalue j is the eigenvalue of the jth principal component, variance_of_original_variable j is the variance of the jth original variable.

[0159] By transposing the factor loading matrix, we get the weight matrix W, which represents the weight of each original feature in each principal component. We calculate the covariance matrix of the data, and by performing eigenvalue decomposition on the covariance matrix, we get the eigenvalues ​​and eigenvectors. Specifically, the contribution of each principal component to the total variance explained by the principal component can be calculated by the following formula:

[0160]

[0161] Among them, λ i is the i-th eigenvalue, n is the number of original variables, and we get the vector V = (v 1 ,v 2 ,…,v p )

[0162] Correspondingly, the dot product between the weight matrix W and the variance contribution is calculated to obtain the weighted value of each principal component, expressed as: dot(W,V).

[0163] Furthermore, the contribution of each original variable to all principal components is normalized by dividing it by the variance contribution ratio of all principal components. The expression is:

[0164]

[0165] Where: list_weight i Represents the feature weight of the i-th original variable. var_weight ik represents the factor loading of the ith original variable on the kth principal component. k represents the explained variance ratio of the kth principal component. n represents the number of original variables and m represents the number of principal components.

[0166] Finally, the obtained similarity set is summed according to its feature weights to obtain the similarity of the comprehensive index. Since the weighting method can be obtained by combining the weighted average formula in the above embodiment, it will not be repeated here.

[0167] Step S14: Indicator duplication identification module, for details, please refer to Figure 6, based on the similarity of basic information elements, description information elements, logical structure elements, and comprehensive similarity of each group of indicators, identify the similarity between the new indicators and the existing indicators: if there are existing indicators with similarity greater than the set threshold K1 in any of the basic information elements similarity TOPN, description information elements similarity TOPN, logical structure elements similarity TOPN, and comprehensive similarity TOPN, then the new indicator is determined as a duplicate indicator; if the similarity TOP1 of the above four aspects is less than the threshold K2, then the new indicator is determined as a non-duplicate indicator and added to the indicator library to enrich the enterprise's indicator system; if none of the above conditions are met, then the stock indicators ranked TopN in basic information element similarity, TopN in description information element similarity, TopN in logical structure element similarity, and TopN in comprehensive similarity under the indicator library are output for developers to provide indicator development suggestions.

[0168] This technical solution combines the standardization of indicator definitions, multiple similarity calculation technologies, multi-dimensional similarity fusion calculation technologies and other methods to automatically calculate the similarity between new indicators and existing indicators, thereby saving time and human resources. Secondly, through the indicator duplication identification module, enterprises can identify and exclude duplicate indicators early, reduce errors and inconsistencies, and help ensure the value of indicators in the indicator library and improve the quality of indicators in the indicator library.

[0169] In the production and operation process of an enterprise, each department needs to define its own business indicators according to the annual goals every year in order to monitor the performance of each department, help the management make decisions, and promote the high-quality development of the company. However, due to the lack of standardization and departmental collaboration, the problem of duplicate indicators has emerged. The duplicate indicators not only waste time and resources, but also increase management and maintenance costs. The existing methods for identifying indicator similarity by communication operators are mainly based on business experience, which relies on the professional knowledge and experience of employees. Employees compare the indicator data of different companies to determine the similarity between them. This method can identify similar indicators to a certain extent, but it also has some obvious shortcomings. For example, employees have different professional knowledge, and different employees may make inconsistent judgments and decisions when comparing the same indicator group; on the other hand, the existing methods require enterprises to invest a lot of human resources, which further increases the human resource costs of enterprises. The existing methods for judging the similarity of indicators are based on the properties of polynomial rings or the decomposition logic of indicators, converting the original indicators into standard calculation formulas or a combination of their internal basic elements. The existing methods are too dependent on the types of indicators, the similarity calculation methods are single and the evaluation is incomplete, with calculation limitations and implementation complexity, and they do not make full use of the indicator text information, which makes them unable to adapt to the complex and changeable enterprise operation needs, and cannot effectively reduce the duplication of indicators, resulting in serious consequences such as waste of enterprise resources. The present invention solves the problems existing in the existing indicator similarity judgment methods through a method of indicator standardization, semantic recognition technology based on deep learning, multi-dimensional similarity calculation, and fusion similarity calculation based on principal component analysis, and improves the accuracy and adaptability of indicator similarity judgment.

[0170] The present invention provides a technology for identifying duplicate indicator construction, which can realize the similarity calculation of new telecommunications business indicators and existing business indicators, automatically identify similar indicators, and help the group company to improve its post-evaluation risk prevention and control system and enhance the company's data governance capabilities.

[0171] The technical points and corresponding beneficial effects of the present invention include:

[0172] 1. The existing indicator similarity judgment technology has a relatively complex indicator disassembly, and the indicator disassembly results are uncertain and the disassembly steps are cumbersome. This technical solution establishes a set of clearly defined indicator specification systems to solve the problem of complex indicator disassembly in existing methods. By dividing and analyzing the elements of each indicator, the indicator disassembly steps are simplified, reducing the complexity and uncertainty of the disassembly process.

[0173] 2. The existing indicator similarity judgment technology is too dependent on the type of indicators, has fewer evaluation dimensions, and a single similarity calculation method, which may lead to limited applicability and poor accuracy of the method, and thus fail to effectively solve the problem of indicator duplication. In response to the above problems, this technical solution adopts a multi-dimensional similarity calculation method. Combining the calculation of the similarity of the basic information elements of the indicators, the similarity of the descriptive information elements, the similarity of the logical structure elements, and the similarity of the comprehensive aspects, the similarity between the indicators is more comprehensively evaluated, making the similarity judgment more flexible and accurate.

[0174] The present invention is not limited to providing similarity calculations for new indicators and existing indicators, but is also applicable to application scenarios such as legal compliance inspections of telecommunication operation indicators, guidance on decision-making for telecommunication product improvements, and expansion and development of new indicators. In addition, the present invention is not limited to the operator industry, and the identification of duplicate construction of risk indicators, performance indicators, supply chain production link indicators, product performance indicators, etc. in other industries or organizations can be solved by the present invention.

[0175] On the basis of the method provided above, the embodiment of this specification also provides a device for identifying duplicate construction of indicators based on text similarity, the device comprising:

[0176] A specification standard establishment module is used to establish specification standards for repeated identification of indicator construction based on the stock indicators in the indicator library. The specification standards are used to represent the similarity calculation rules of the specified element combination in the application field. The specified element combination includes at least basic information elements, descriptive information elements and logical structure elements.

[0177] An element data acquisition module is used to acquire a newly added indicator and obtain element data corresponding to a specified element combination of the newly added indicator;

[0178] A similarity calculation module, used to obtain similarity results between the new indicator and the existing indicator on different elements according to similarity calculation rules of different elements and element data corresponding to the specified element combination of the new indicator;

[0179] A factor loading matrix acquisition module is used to obtain the factor loading matrix of different elements through principal component analysis according to the similarity result;

[0180] A comprehensive aspect similarity calculation module, used to obtain the comprehensive aspect similarity between the newly added indicator and each existing indicator according to the factor loading matrix;

[0181] A screening module is used to obtain the similarity ranking top 10 of the new indicator and the existing indicator in basic information elements, description information elements, logical structure elements, and comprehensive similarity top 10, where N≥3;

[0182] The first judgment module determines that the newly added indicator is a duplicate indicator when there are indicators in the similarity ranking TOPN whose basic information elements, description information elements, logical structure elements and comprehensive similarities are all greater than a first preset threshold.

[0183] The second judgment module is used to determine that the newly added indicator is a non-repetitive indicator and add it to the indicator library when each similarity ranking TOP1 in all the similarity ranking TOPN is less than a second preset threshold, wherein the second preset threshold is less than the first preset threshold.

[0184] This embodiment provides a computer device, whose internal structure diagram can be shown as follows: Figure 7 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection.

[0185] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0186] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0187] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0188] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0189] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0190] It should also be understood that in the embodiments of this article, the term "and / or" is only a description of the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0191] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this article.

[0192] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0193] In the several embodiments provided herein, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or can be electrical, mechanical or other forms of connection.

[0194] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of this article.

[0195] Specific embodiments are used in this article to illustrate the principles and implementation methods of this article. The description of the above embodiments is only used to help understand the methods and core ideas of this article. At the same time, for general technicians in this field, according to the ideas of this article, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on this article.

Claims

1. A method for identifying duplicate construction of indicators based on text similarity, characterized in that: The method comprises: According to the existing indicators in the indicator library, establish the specification standard for repeated identification of indicator construction, the specification standard is used to represent the similarity calculation rules of the specified element combination in the application field, and the specified element combination at least includes basic information elements, descriptive information elements and logical structure elements; Obtain a newly added indicator, and obtain element data corresponding to a specified element combination of the newly added indicator; According to the similarity calculation rules of different elements and the element data corresponding to the specified element combination of the newly added indicator, the similarity results between the newly added indicator and the existing indicator on different elements are obtained; According to the similarity results, the factor loading matrix of different elements is obtained through principal component analysis; According to the factor loading matrix, the comprehensive aspect similarity between the newly added indicator and each existing indicator is obtained; Obtain the similarity ranking top 10 of the new indicator and the existing indicator in basic information elements, description information elements, logical structure elements, and comprehensive similarity top 10, where N≥3; When there are indicators in the similarity ranking TOPN whose basic information elements, description information elements, logical structure elements and comprehensive aspects are all greater than the first preset threshold, the newly added indicator is determined to be a duplicate indicator; When each similarity ranking TOP1 in all the similarity ranking TOPN is less than a second preset threshold, the newly added indicator is determined to be a non-repetitive indicator and is added to the indicator library, wherein the second preset threshold is less than the first preset threshold.

2. The method according to claim 1, characterized in that Based on the existing indicators in the indicator database, establish the standard for repeated identification of indicator construction, including: Determine the application areas corresponding to the indicator library; Determine, according to the application field, the indicator-related information in the indicator library, wherein the indicator-related information at least includes indicator definition, calculation logic and generation method; According to the indicator-related information, a specified element combination corresponding to the indicator in the indicator library is determined, and a similarity calculation rule for each element in the indicator element combination is established.

3. The method according to claim 1, characterized in that The similarity calculation process for basic information elements is as follows: Determine the basic information type corresponding to the basic information element of the indicator in the indicator library; According to the basic information type, the newly added indicator and the existing indicator are respectively subjected to word segmentation processing to obtain word segmentation results of the newly added indicator and the existing indicator; According to the segmentation results of the newly added indicator and the existing indicator, the similarity results of the newly added indicator and the existing indicator in the basic information elements are obtained by the following formula; Among them, S(I_new.types) is the word segmentation result of the basic information elements in the new indicators, S(I_i.types) is the word segmentation result of the basic information elements in the existing indicators, and Sim is the similarity result between the new indicators and the existing indicators in the basic information elements.

4. The method according to claim 1, characterized in that: The similarity calculation process for the description information elements is as follows: Performing word segmentation processing on the description information elements of the newly added indicator and the existing indicator respectively to obtain word segmentation tag sequences of the newly added indicator and the existing indicator; Input the word segmentation token sequences into the ALBERT pre-training model respectively to obtain an embedding vector sequence corresponding to each word segmentation token sequence; Input the embedding vector sequences of the newly added indicators and the existing indicators into the Transformer encoder in the ALBERT model respectively to obtain the encoded embedding vector sequence, and use the first embedding vector in the top-level encoding result in the Transformer encoder as the target embedding vector; According to the target embedding vector of the newly added indicator and the target embedding vector of the existing indicator, the similarity between the added indicator and the existing indicator in the description information element is calculated.

5. The method according to claim 1, characterized in that: The similarity calculation process for logical structure elements is as follows: When the logical structure element is a logical structure text, a standardized element text is extracted from the logical structure text using a text generation technology, wherein the text generation technology at least includes classifying and parsing the logical structure text; Performing character conversion, cleaning, and segmentation steps on the standardized element text to obtain target text elements; According to the target text elements of the newly added indicator and the existing indicator, the similarity between the newly added indicator and the existing indicator in the logical structure elements is calculated by using the cosine distance calculation method; When the logical structure element is a mathematical calculation formula, the variables and mathematical symbols in the mathematical calculation formula are determined; For the variables given in the formula, the Word2Vec technology is used to calculate the semantic similarity of the variables between formulas; With respect to the mathematical symbols in the formula, by comparing the algebraic expressions, symbol usage and structural composition in the formula, the similarity between the newly added indicators and the existing indicators in terms of logical structural elements is obtained.

6. The method according to claim 1, characterized in that According to the similarity results, the factor loading matrix of different elements was obtained through principal component analysis, which also included: Obtaining the similarity weight of each element of the newly added indicator and each of the existing indicators in the specified element combination; According to the similarity weight, and by obtaining the similarity between the newly added indicator and each of the existing indicators in each element of the specified element combination, the final similarity between the newly added indicator and the existing indicators on each element in the specified element combination is obtained.

7. The method according to claim 6, characterized in that According to the similarity result, the factor loading matrix of different elements is obtained through principal component analysis, including: Obtaining a similarity vector between the newly added indicator and the existing indicators according to the similarity between each element in the specified element combination and the newly added indicator; Determine, according to the similarity vector, a representative factor corresponding to each similarity by using a principal component analysis method, wherein the representative factor represents a feature weight of each similarity; The factor loading matrix is ​​determined according to the representative factors corresponding to each similarity.

8. The method according to claim 7, characterized in that According to the factor loading matrix, the comprehensive similarity between the new indicators and the existing indicators is obtained, including: The final similarity between the newly added indicator and the existing indicator on each element in the specified element combination, and the factor loading matrix, obtain the comprehensive aspect similarity between the newly added indicator and the existing indicator.

9. The method according to claim 1, characterized in that: The method further comprises: When there is at least one similarity among all the similarity rankings TOPN that is less than or equal to the first preset threshold, and there is at least one similarity ranking TOP1 among all the similarity rankings TOPN that is greater than or equal to the second preset threshold, the stock indicator list corresponding to the similarity rankings TOPN of the basic information elements, the stock indicator list corresponding to the similarity rankings TOPN of the descriptive information elements, and the stock indicator list corresponding to the similarity rankings TOPN of the logical structure elements are summarized to obtain a candidate stock indicator list, so as to perform manual judgment based on the candidate stock indicator list.

10. A device for identifying duplicate constructions based on text similarity indicators, characterized in that: The device comprises: A specification standard establishment module is used to establish specification standards for repeated identification of indicator construction based on the stock indicators in the indicator library. The specification standards are used to represent the similarity calculation rules of the specified element combination in the application field. The specified element combination includes at least basic information elements, descriptive information elements and logical structure elements. An element data acquisition module is used to acquire a newly added indicator and obtain element data corresponding to a specified element combination of the newly added indicator; A similarity calculation module, used to obtain similarity results between the new indicator and the existing indicator on different elements according to similarity calculation rules of different elements and element data corresponding to the specified element combination of the new indicator; A factor loading matrix acquisition module is used to obtain the factor loading matrix of different elements through principal component analysis according to the similarity result; A comprehensive aspect similarity calculation module, used to obtain the comprehensive aspect similarity between the newly added indicator and each existing indicator according to the factor loading matrix; A screening module is used to obtain the similarity ranking top 10 of the new indicator and the existing indicator in basic information elements, description information elements, logical structure elements, and comprehensive similarity top 10, where N≥3; A first judgment module is used to determine that the newly added indicator is a duplicate indicator when there are indicators in the similarity ranking TOPN whose basic information elements, description information elements, logical structure elements and comprehensive aspects are all greater than a first preset threshold; The second judgment module is used to determine that the newly added indicator is a non-repetitive indicator and add it to the indicator library when each similarity ranking TOP1 in all the similarity ranking TOPN is less than a second preset threshold, wherein the second preset threshold is less than the first preset threshold.

Images