A method for judging the consistency of unit names

Abstract

A unit name consistency discrimination method, comprising: constructing a standard name library, dividing each standard unit name and setting a category; obtaining a user input name, dividing the input name and setting a category; constructing a consistency judgment library; constructing a combined name set for each standard unit name in the consistency judgment library; calculating the cosine similarity of the input name and each standard unit name, and selecting the maximum cosine similarity and the corresponding standard unit name therefrom; using a sequence matching method, calculating the sequence similarity of the input name and each standard unit name, and selecting the maximum sequence similarity and the corresponding standard unit name therefrom; judging whether the standard unit name with the maximum cosine similarity and the standard unit name with the maximum sequence similarity are consistent, and if so, the standard unit name is consistent with the user input name. The present application belongs to the field of information, and can accurately identify, match and obtain a standard unit name consistent with the user input name.

Description

Technical Field

[0001] This invention relates to a method for determining the consistency of unit names, belonging to the field of information. Background Technology

[0002] In various business scenarios in banking, securities, and insurance, it is necessary to accurately obtain the company information filled in by users. However, this company data is manually filled in by users and often does not conform to the standard names registered with the business administration. Incorrect or missing information from users, as well as diverse abbreviations, often make it difficult to correctly identify the company name.

[0003] Generally, there are the following problems in judging the consistency of unit names:

[0004] (1) Inaccurate names. The unit data is manually filled in by the user, and there are often problems such as missing or incorrect unit names, use of inconsistent unit abbreviations, and failure to distinguish between branch units and superior units.

[0005] (2) Diverse abbreviations. Most abbreviations of organizations are often formed by people according to custom and there is no standard form. They include Chinese pinyin, abbreviations, former names, transliterations, etc.

[0006] Patent application CN 202111505876 (Application title: A Company Information Matching Method, Applicant: Beijing Yule Innovation Technology Co., Ltd., Application date: 2021.12.10) discloses a company information matching method. The method includes: receiving company information to be searched input by a user; formatting the phone number of the company to be searched; performing a fuzzy search on the name of the company to be searched to obtain a first result set; the first result set includes at least one similar company information retrieved; determining whether the name of the company to be searched is consistent with the similar company name in the first result set; if the company names are consistent, determining whether the phone number of the similar company name corresponding to the similar company name consistent with the name of the company to be searched is consistent with the formatted phone number of the company to be searched; if the phone numbers are consistent, determining that the matching result of the company information to be searched is consistent, wherein the company information includes the company name, the company phone number, and the company address. This technical solution determines whether company information matches using company name and phone number. When judging whether the string of the company name to be searched is equal to the string of the currently similar company name, it adopts the form of "region + brand name + industry name + enterprise type" based on experience and human-defined rules. This lacks universality and only judges based on individual strings, ignoring the relationship between strings, which can easily lead to confusion in the matching results.

[0007] Therefore, how to accurately identify the unit name entered by the user and thus match it with a standard unit name that matches the user's input has become a key technical issue of concern for technicians. Summary of the Invention

[0008] In view of this, the purpose of the present invention is to provide a method for determining the consistency of unit names, which can accurately identify the unit names entered by the user, thereby matching and obtaining a standard unit name that is consistent with the user-entered name.

[0009] To achieve the above objectives, the present invention provides a method for determining the consistency of unit names, comprising:

[0010] Step 1: Construct a standard name library, segment each standard unit name in the standard name library into words, and set the category of each word segment. The categories of word segmentation include region (R), keyword (X), industry (I), and unit suffix (O).

[0011] Step 2: Obtain the user-input name, segment the user-input name into words, and set the category of each word;

[0012] Step 3: Construct a consistency judgment library based on the standard name library;

[0013] Step 4: Construct a combined name set for each standard unit name in the consistency judgment library: Combine the word segments of each standard unit name according to their categories to generate a new combined name. The multiple combined names obtained by combining them according to various categories constitute the combined name set for each standard unit name.

[0014] Step 5: Calculate the cosine distance between the user input name and each combination name in the consistency judgment database, then calculate the weighted cosine similarity between the user input name and each standard unit name, and finally select the maximum cosine similarity and the corresponding standard unit name.

[0015] Step 6: Using the sequence matching method, calculate the sequence matching value of each combination name in the combination name set of each standard unit name in the consistency judgment database, then calculate the sequence similarity between the user input name and each standard unit name by weighting, and finally select the maximum sequence similarity and the corresponding standard unit name.

[0016] Step 7: Determine whether the standard unit name corresponding to the maximum cosine similarity is consistent with the standard unit name corresponding to the maximum sequence similarity. If they are consistent, the standard unit name corresponding to the maximum cosine similarity is consistent with the name entered by the user, and the maximum sequence similarity is saved as the consistency discrimination value. This process ends. If not, continue to the next step.

[0017] Step 8: Calculate the sequence matching values ​​of the standard unit name and the user input name corresponding to the maximum cosine similarity and the maximum sequence similarity, respectively. Then select the maximum value, where the standard unit name corresponding to the maximum value is the same as the user input name, and save the maximum value as the consistency discrimination value.

[0018] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention designs a dedicated word segmentation method for short texts of unit names. First, the unit name is segmented into words. Then, the cosine distance and Sequence Matcher methods are integrated. Based on this, different weights are assigned to word vectors for weighted calculation to output the final score. Finally, for user input names with relatively ambiguous discrimination results, unsupervised learning is used to model the data and output the final score, thereby accurately matching and obtaining standard unit names consistent with user input names. When segmenting unit names, this invention uses the NLPIR library of the Chinese Academy of Sciences for word segmentation and part-of-speech tagging. Then, based on the part-of-speech of each part, the type is set, and the parts are arranged and combined. During calculation, the various combinations are weighted and calculated as the matching score, thereby better preserving the correlation between strings and accurately identifying diverse abbreviations. Attached Figure Description

[0019] Figure 1 This is a flowchart of a method for determining the consistency of unit names according to the present invention.

[0020] Figure 2 yes Figure 1 Step five involves calculating the cosine distance between the combined names in the set of combined names of the user-input name and the standard unit name. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.

[0022] like Figure 1 As shown, the present invention provides a method for determining the consistency of unit names, comprising:

[0023] Step 1: Construct a standard name library, segment each standard unit name in the standard name library into words, and set the category of each word segment. The categories of word segmentation include region (R), keyword (X), industry (I) and unit suffix (O).

[0024] Step 2: Obtain the user-input name, segment the user-input name into words, and set the category of each word;

[0025] Step 3: Based on the standard name library, construct a consistency judgment library: Extract the first character S of the word segment with category X in the user input name, and then judge whether each standard unit name in the standard name library contains a word segment with category X and first character S. If so, write the standard unit name into the consistency judgment library.

[0026] Extracting all data from the full standard name library where the first character of keyword X is the same as the first character of keyword X in the user-input name, and using this data as the consistency judgment library for this judgment, can significantly reduce the size of the standard library and improve efficiency and accuracy.

[0027] Step 4: Construct a combined name set for each standard unit name in the consistency judgment library: Combine the word segments of each standard unit name according to their categories to generate a new combined name. The multiple combined names obtained by combining them according to various categories constitute the combined name set for each standard unit name.

[0028] Step 5: Calculate the cosine distance between the user input name and each combination name in the consistency judgment database, then calculate the weighted cosine similarity between the user input name and each standard unit name, and finally select the maximum cosine similarity and the corresponding standard unit name.

[0029] Step 6: Using the sequence matching method, calculate the sequence matching value of each combination name in the combination name set of each standard unit name in the consistency judgment database, then calculate the sequence similarity between the user input name and each standard unit name by weighting, and finally select the maximum sequence similarity and the corresponding standard unit name.

[0030] Step 7: Determine if the standard unit name corresponding to the maximum cosine similarity is consistent with the standard unit name corresponding to the maximum sequence similarity. If yes, the standard unit name corresponding to the maximum cosine similarity is consistent with the user-input name, and the maximum sequence similarity is saved as the consistency judgment value, and this process ends; if no, continue to the next step.

[0031] Step 8: Calculate the sequence matching values ​​of the standard unit name and the user input name corresponding to the maximum cosine similarity and the maximum sequence similarity, respectively. Then select the maximum value, where the standard unit name corresponding to the maximum value is the same as the user input name, and save the maximum value as the consistency discrimination value.

[0032] This invention can extract unit name data from the standard name table for business registration and the organization name table as standard unit names to construct a standard name library, and perform data preprocessing on the user input names in the user input table and the standard unit names in the standard name library. The specific steps are as follows:

[0033] 1) Filter out data with empty unit names from the user input table, the standard name table for business registration, and the organization name table;

[0034] 2) Remove special characters from the text, such as “?”, “^”, “(”, “)”, “《”, “》”, etc.

[0035] 3) Data consisting entirely of numbers and letters is considered dirty data and processed accordingly;

[0036] 4) Remove leading, trailing, and middle spaces from the text;

[0037] 5) Replace full-width and half-width Arabic numerals in the text with Chinese numerals.

[0038] Create a user dictionary from pre-prepared unit name suffixes and keywords, and import it into the word segmentation environment. To improve the accuracy of identification in industries with many branches, such as banks and insurance companies, remove stop words in advance, such as words that affect the identification results, such as "branch," "sub-branch," and "joint-stock company."

[0039] Since a company name typically consists of four parts: region (R), keyword (X), industry (I), and company suffix (O), this invention uses the RXIO method to represent the company name structure. In step one or step two, setting the category of each word segment can further include:

[0040] The NLPIR Chinese word segmentation system is used to segment and tag the parts of speech of standard unit names or user-input names. The parts of speech include: location, noun, verb, and adverb. Then, the category of each segment is set one by one: if the part of speech is location, the category of the segment is region R; if the part of speech is noun and the segment exists in the unit suffix thesaurus, the category of the segment is unit suffix O; if the part of speech is noun, verb, or adverb and the segment exists in the industry thesaurus, the category of the segment is industry I; if the segment exists in the unit keyword thesaurus, or does not exist in any thesaurus and does not belong to the above three categories, the category of the segment is keyword X.

[0041] Experiments have shown that step four utilizes three category combinations for optimal technical performance: keyword X, location R + keyword X, and keyword X + industry I. Taking Shanghai Worison Fitness Company as an example, its combined name set could be {Worison, Shanghai Worison, Worison Fitness}. The following is an example of segmenting and categorizing a standard company name or a user-input name:

[0042] name Word segmentation X R+X X+I Xi'an **Jia Water Treatment Technology Co., Ltd. {'Water Treatment', 'Limited Company', '**Jia', 'Xi'an', 'Technology'} **Jia Xi'an **Jia **Jia Water Treatment *Mountain Scenic Area {'*mountain', 'scenery', 'scenic spot'} landscape Mountain scenery Scenic spots Tianjin *Jian Culture Communication Co., Ltd. {'Cultural Communication', 'See', 'Limited Company', '*', 'Tianjin'} *See Tianjin* *See Cultural Communication Jiangsu ** Landscape Architecture Co., Ltd. {'Architecture', 'Limited Company', '**', 'Jiangsu', 'Landscape Architecture'} **garden Jiangsu ** Garden **Garden Architecture Shandong Shuo Automation Co., Ltd. {'Limited Company', '*Master', 'Automation', 'Shandong'} *large Shandong* Master *Shuo Automation

[0043] like Figure 2 As shown, in step five, the user-input name is calculated. and standard unit name Combination names in the set of combination names The cosine distance can further include:

[0044] Step 51: Input the user's name All corresponding word segments constitute the user word segmentation set. ;

[0045] Step 52: Combine the names All corresponding word segments constitute a combined word segmentation set. ;

[0046] Step 53: Analyze the user-segmented word set and combined word segmentation set Merge into a general vocabulary set and remove Repeated word segmentation in;

[0047] Step 54: Input the user's name Convert into user word vectors Create a set of length and total words. The user word vectors with the same number of elements are initialized to 0, and then the total word set is checked one by one. Does each element in exist in If yes, then modify the corresponding position of the user's word vector to 1; otherwise, continue the judgment. The next element in;

[0048] Step 55: Combine the names Transform into combined word vectors Create a set of length and total words. Combine word vectors with the same number of elements and initialize them to 0. Then, check the total word set one by one. Does each element in exist in If yes, then change the corresponding position of the combined word vector to 1; otherwise, continue the judgment. The next element in;

[0049] Step 56: Calculate user word vectors and combined word vectors The cosine distance is calculated using the following formula: = 1 - cos(A,B) = .

[0050] Based on industry experience and extensive data testing, it was found that because place names, industries, and organizational structures all have some degree of repetition, while keyword X is relatively unique, keyword X has a significant impact on the accuracy of the judgment results. Therefore, keyword X is assigned the highest weight. The influence of the combination of location R + keyword X and keyword X + industry I is secondary, and their importance is comparable. Therefore, they are assigned a smaller but equal weight than keyword X. Thus, when using the three combination methods—keyword X, location R + keyword X, and keyword X + industry I—the optimal technical effect is achieved when keyword X has the highest weight in the weighted calculations of steps five and six, and the weights of location R + keyword X and keyword X + industry I are equal.

[0051] In step five, the formula for calculating the weighted cosine similarity between the user-input name and each standard unit name is as follows: ,in, User input name and standard unit name cosine similarity, It is the cosine distance of the i-th combined name in the set of combined names of user input names and standard unit names. is the weight value of the i-th category combination method, and M is the number of combination names in the combination name set.

[0052] Sequence matching is a common method in short text matching. For sequences of the same length, the distance between any two points is calculated and then summed; the smaller the distance, the higher the similarity. For sequences of different lengths, subsequence matching and sliding windows can be used. To improve efficiency and accuracy, this invention uses sequence matching for similarity calculation in specific industries. Before starting the discriminant analysis, unweighted sequence matching is performed on industry units with many subordinate branches (such as banks and insurance companies), as well as public institutions and government agencies, within their respective institutional categories to reduce subsequent computation. Verification has shown that this method is more accurate for such industries. Specifically, more detailed rules are set for banks down to the name level to increase the accuracy of bank name discrimination and avoid errors in bank name discrimination when performing industry discrimination in a broad category standard library. In step six, the formula for weighted calculation of the sequence similarity between the user-input name and each standard unit name is as follows: ,in, User input name and standard unit name Sequence similarity, It is the sequence match value of the i-th combined name in the set of combined names of user input name and standard unit name. is the weight value of the i-th category combination method, and M is the number of combination names in the combination name set.

[0053] In step seven or eight, if the consistency discrimination value is lower than a set threshold, it may also include:

[0054] Step 9: Using Word2Vec, construct word vectors corresponding to the user input name and all standard unit names in the consistency judgment library. Then, input the word vectors of the user input name and all standard unit names in the consistency judgment library into the unsupervised learning KD-tree algorithm, and output the standard unit name closest to the user input name and the distance score. The output standard unit name is the unit name consistent with the user input name, and the distance score is saved as the consistency judgment value.

[0055] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for determining the consistency of unit names, characterized in that, Including: Step 1: Construct a standard name library, segment each standard unit name in the standard name library into words, and set the category of each word segment. The categories of word segmentation include region (R), keyword (X), industry (I), and unit suffix (O). Step 2: Obtain the user-input name, segment the user-input name into words, and set the category of each word; Step 3: Construct a consistency judgment library based on the standard name library; Step 4: Construct a combined name set for each standard unit name in the consistency judgment library: Combine the word segments of each standard unit name according to their categories to generate a new combined name. The combined names obtained by combining them in multiple categories constitute the combined name set for each standard unit name. The multiple category combination methods are: keyword X, location R + keyword X, and keyword X + industry I. Step 5: Calculate the cosine distance between the user input name and each combination name in the consistency judgment database, then calculate the weighted cosine similarity between the user input name and each standard unit name, and finally select the maximum cosine similarity and the corresponding standard unit name. Step 6: Using the sequence matching method, calculate the sequence matching value of each combination name in the combination name set of each standard unit name in the consistency judgment database, then calculate the sequence similarity between the user input name and each standard unit name by weighting, and finally select the maximum sequence similarity and the corresponding standard unit name. Step 7: Determine whether the standard unit name corresponding to the maximum cosine similarity is consistent with the standard unit name corresponding to the maximum sequence similarity. If they are consistent, the standard unit name corresponding to the maximum cosine similarity is consistent with the name entered by the user, and the maximum sequence similarity is saved as the consistency discrimination value. This process ends. If not, continue to the next step. Step 8: Calculate the sequence matching values ​​of the standard unit name and the user input name corresponding to the maximum cosine similarity and the maximum sequence similarity, respectively. Then select the maximum value, where the standard unit name corresponding to the maximum value is the same as the user input name, and save the maximum value as the consistency discrimination value.

2. The method according to claim 1, characterized in that, In step one or step two, the category of each word segment is set, which further includes: The standard unit name or user-input name is segmented and labeled with part-of-speech tags. The part-of-speech tags include: location, noun, verb, and adverb. Then, the category of each segment is set one by one: if the part-of-speech tag is location, the category of the segment is region R; if the part-of-speech tag is noun and the segment exists in the unit suffix dictionary, the category of the segment is unit suffix O. If the part of speech is a noun, verb, or adverb, and the word segment exists in the industry thesaurus, then the word segment category is Industry I; if the word segment exists in the unit keyword thesaurus, or does not exist in any thesaurus and does not belong to the above three categories, then the word segment category is Keyword X.

3. The method according to claim 1, characterized in that, Step three further includes: Extract the first character S of the word segment with category X in the user input name, and then check whether each standard unit name in the standard name library contains a word segment with category X and first character S. If so, write the standard unit name into the consistency judgment library.

4. The method according to claim 1, characterized in that, In step five, calculate the user-input name. and standard unit name Combination names in the set of combination names The cosine distance further includes: Step 51: Input the user's name All corresponding word segments constitute the user word segmentation set. ; Step 52: Combine the names All corresponding word segments constitute a combined word segmentation set. ; Step 53: Analyze the user-segmented word set and combined word segmentation set Merge into a general vocabulary set and remove Repeated word segmentation in; Step 54: Input the user's name Convert into user word vectors Create a set of length and total words. The user word vectors with the same number of elements are initialized to 0, and then the total word set is checked one by one. Does each element in exist in If yes, then modify the corresponding position of the user's word vector to 1; otherwise, continue the judgment. The next element in; Step 55: Combine the names Transform into combined word vectors Create a set of length and total words. Combine word vectors with the same number of elements and initialize them to 0. Then, check the total word set one by one. Does each element in exist in If yes, then change the corresponding position of the combined word vector to 1; otherwise, continue the judgment. The next element in; Step 56: Calculate user word vectors and combined word vectors The cosine distance is calculated using the following formula: = 1 - cos(A,B) = .

5. The method according to claim 1, characterized in that, When using the three combination methods of keyword X, location R + keyword X, and keyword X + industry I, in the weighted calculation of steps five and six, the weight value of keyword X is greater than the weight values ​​of location R + keyword X and keyword X + industry I, and the weight values ​​of location R + keyword X and keyword X + industry I are equal.

6. The method according to claim 1, characterized in that, In step seven or eight, if the consistency discrimination value is lower than the set threshold, the following steps are also included: Step 9: Using Word2Vec, construct word vectors corresponding to the user input name and all standard unit names in the consistency judgment library. Then, input the word vectors of the user input name and all standard unit names in the consistency judgment library into the unsupervised learning KD-tree algorithm, and output the standard unit name closest to the user input name and the distance score. The output standard unit name is the unit name consistent with the user input name, and the distance score is saved as the consistency judgment value.

Images