CN114911999A - A name matching method and device - Google Patents

Abstract

Embodiments of the present invention provide a name matching method and device, which relate to the field of big data. The method includes: in response to a search request for an original name to be searched, matching the original name based on characters; if the matching fails, Then, split the original name to obtain a plurality of word segments, and classify the plurality of word segments according to preset categories to obtain a plurality of word segments with categories; Recombining each word segment to obtain a plurality of candidate names; match the plurality of candidate names with the preset name database for similarity respectively, and determine the target candidate name with the highest similarity; The candidate names are semantically classified and matched with the name database to obtain a matching result. The embodiment of the present invention reduces the workload of matching calculation and improves the matching accuracy.

Description

A name matching method and device

technical field

The invention relates to the technical field of big data, in particular to a name matching method and a name matching device.

Background technique

In order to achieve all-round customer insight and product recommendation, the relevant technical support obtains the corresponding customer information by entering the company name in the system, including customer development information, income information, bidding information, business information, enterprise map and complaint information, etc. Connect the company name (the company name obtained from the web page) and the customer name (the company name recorded and stored in the system) to get all the information corresponding to the customer name in the system. However, current users often enter name abbreviations/abbreviations when entering company names. Due to irregular input, typos, complex and diverse abbreviations, etc., the system cannot accurately find the customer name.

In order to solve the above problems, the solution in the related art is usually to perform similarity matching between the company name and the customer name, sort all the matching scores, and obtain the customer name whose matching score is higher than a certain threshold as the matching result of the company name. The matching accuracy of this similarity matching method is usually low.

In addition, the similarity algorithm of the shortest edit distance is generally used when calculating string matching. It considers the semantic relationship between text contexts as a whole and is a commonly used distance function measurement method. In the field of string similarity matching, it is obtained It has been widely used, but there are still some problems:

1) The traditional edit distance algorithm only considers the number of editing operations and does not have universal applicability;

2) The traditional edit distance algorithm has a certain deviation in the calculation of long string insertion and deletion errors, resulting in low matching accuracy.

SUMMARY OF THE INVENTION

In view of the above problems, the embodiments of the present invention are proposed to provide a name matching method and a corresponding name matching apparatus that overcome the above problems or at least partially solve the above problems.

In order to solve the above problem, an embodiment of the present invention discloses a name matching method, characterized in that the method includes:

in response to a search request for the original name to be searched, matching the original name based on the characters;

If the matching fails, the original name is split to obtain a plurality of word segments, and the plurality of word segments are classified according to preset categories to obtain a plurality of word segments with categories;

Recombining the plurality of word segmentations based on the target category in the category to obtain a plurality of candidate names;

Perform similarity matching between the multiple candidate names and a preset name database respectively, and determine the target candidate name with the highest similarity;

Based on the trained classification model, semantic classification matching is performed between the target candidate name and the name database, and a matching result is obtained.

Preferably, the character-based matching of the original name includes:

detecting whether a name with the same characters as the original name exists in the name database;

If it exists, the match succeeds; if it does not exist, the match fails.

Preferably, the original name is split to obtain a plurality of word segments, and the plurality of word segments are classified according to a preset type to obtain a plurality of word segments with categories, including:

Use jieba to split the original name to obtain multiple participles;

Match each participle with a preset category library, determine the one-to-one corresponding category of each participle, and obtain multiple participles with categories.

Preferably, the multiple word segmentations are reorganized based on the target category in the category to obtain multiple candidate names, including:

Recombining the word segments with the target category and the word segments with the non-target category among the plurality of word segmentations to obtain a plurality of candidate names.

Preferably, the similarity matching is performed on the multiple candidate names with a preset name database respectively, and the target candidate name with the highest similarity is determined, including:

Carrying out similarity matching between the multiple candidate names and the name database respectively, and determining the candidate similarity of each candidate name;

The target candidate similarity with the highest similarity among the plurality of candidate similarities is determined, and the candidate name corresponding to the target candidate similarity is used as the target candidate name.

Preferably, performing similarity matching between the multiple candidate names and the name database, and determining the candidate similarity of each candidate name, including:

For any candidate name in the plurality of candidate names, perform similarity matching between the any candidate name and at least one preset name in the name database to obtain at least one similarity;

A candidate similarity with the highest similarity among the at least one similarity is determined.

Preferably, performing similarity matching between any candidate name and at least one preset name in the name database to obtain at least one similarity, including:

For any preset name in the at least one preset name, obtain the forward maximum common substring and the backward maximum common substring of the any candidate name and the any preset name;

Calculate the forward similarity based on the forward maximum common substring, and calculate the backward similarity using the backward maximum common substring;

Based on the forward similarity and the backward similarity, the similarity between any candidate name and any preset name is calculated.

Preferably, the target candidate similarity with the highest similarity among the multiple candidate similarities is determined, and the candidate name corresponding to the target candidate similarity is used as the target candidate name, including:

Normalize the multiple candidate similarities based on the forward maximum common substring and the backward maximum common substring to obtain the target candidate similarity with the highest similarity;

The candidate name corresponding to the similarity of the target candidate is used as the target candidate name.

Preferably, the classification model based on the training completes semantic classification matching between the target candidate name and the name database to obtain a matching result, including:

Inputting the target candidate name into the trained classification model, so that the classification model uses a preset feature index to perform semantic classification matching between the target candidate name and the name database;

If the matching is successful, the matching preset name will be used as the matching result; if the matching fails, the matching failure information will be generated.

Correspondingly, an embodiment of the present invention discloses a name matching device, characterized in that the device includes:

a first matching module, configured to match the original name based on characters in response to a search request for the original name to be searched;

A word segmentation module, used for splitting the original name if the matching fails, to obtain a plurality of word segmentations, and classifying the plurality of word segmentations according to preset categories to obtain a plurality of word segmentations with categories;

a reorganization module, configured to reorganize the plurality of word segmentations based on the target category in the category to obtain a plurality of candidate names;

The second matching module is used to perform similarity matching between the multiple candidate names and the preset name database respectively, and determine the target candidate name with the highest similarity;

The classification module is configured to perform semantic classification matching on the target candidate name and the name database based on the classification model completed by training, and obtain a matching result.

Preferably, the first matching module is specifically used for:

detecting whether a name with the same characters as the original name exists in the name database;

If it exists, the match succeeds; if it does not exist, the match fails.

Preferably, the word segmentation module is specifically used for:

Use jieba to split the original name to obtain multiple participles;

Match each participle with a preset category library, determine the one-to-one corresponding category of each participle, and obtain multiple participles with categories.

Preferably, the recombination module is specifically used for:

Recombining the word segments with the target category and the word segments with the non-target category among the plurality of word segmentations to obtain a plurality of candidate names.

Preferably, the second matching module includes:

a similarity matching submodule, used to perform similarity matching between the multiple candidate names and the name database respectively, and determine the candidate similarity of each candidate name;

The determination submodule is used to determine the target candidate similarity with the highest similarity among the multiple candidate similarities, and use the candidate name corresponding to the target candidate similarity as the target candidate name.

Preferably, the similarity matching sub-module includes:

a matching unit, configured to perform similarity matching between any candidate name in the plurality of candidate names and at least one preset name in the name database to obtain at least one similarity;

A determining unit, configured to determine a candidate similarity with the highest similarity among the at least one similarity.

Preferably, the matching unit is specifically used for:

For any preset name in the at least one preset name, obtain the forward maximum common substring and the backward maximum common substring of the any candidate name and the any preset name;

Calculate the forward similarity based on the forward maximum common substring, and calculate the backward similarity using the backward maximum common substring;

Based on the forward similarity and the backward similarity, the similarity between any candidate name and any preset name is calculated.

Preferably, the determining submodule is specifically used for:

Normalize the multiple candidate similarities based on the forward maximum common substring and the backward maximum common substring to obtain the target candidate similarity with the highest similarity;

The candidate name corresponding to the similarity of the target candidate is used as the target candidate name.

Preferably, the classification module is specifically used for:

Inputting the target candidate name into the trained classification model, so that the classification model uses a preset feature index to perform semantic classification matching between the target candidate name and the name database;

If the matching is successful, the matching preset name will be used as the matching result; if the matching fails, the matching failure information will be generated.

Correspondingly, an embodiment of the present invention discloses an electronic device, comprising: a processor, a memory, and a computer program stored on the memory and capable of running on the processor, the computer program being executed by the processor Each step of the above name matching method embodiment is implemented.

Correspondingly, an embodiment of the present invention discloses a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, each step of the above name matching method embodiment is implemented.

The embodiments of the present invention include the following advantages:

In response to the search request for the original name to be searched, the background server matches the original name based on the characters; if the matching fails, the original name is split to obtain a plurality of participles, and the multiple participles are divided into two parts. Classify according to preset categories to obtain a plurality of word segmentations with categories, and then reorganize the plurality of word segmentations based on the target categories in the categories to obtain a plurality of candidate names, and combine the plurality of candidate names with the The preset name database performs similarity matching, determines the target candidate name with the highest similarity, and then performs semantic classification matching on the target candidate name and the name database based on the classification model completed to obtain a matching result. In this way, after the exact match of the original name fails, the original name can be disassembled according to the category to obtain multiple word segmentations, and then the multiple segmentations can be reorganized around the specified category in the category to obtain multiple new names, which is easy to operate and reduce The range of word segmentation and combination is reduced, and the workload of matching calculation is reduced. Moreover, for the identified target candidate names, the method of semantic classification and matching is used to convert the traditional similarity matching problem into a machine classification problem, and the similarity of the name pairs is considered from the semantic similarity, which improves the matching accuracy and further reduces the to match the computational workload.

Description of drawings

Fig. 1 is the step flow chart of a kind of name matching method embodiment of the present invention;

FIG. 2 is a structural block diagram of an embodiment of a name matching apparatus according to the present invention.

Detailed ways

In order to make the above objects, features and advantages of the present invention more clearly understood, the present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.

One of the core concepts of the embodiments of the present invention is that, in response to a search request for the original name to be searched, the background server matches the original name based on characters; if the matching fails, splits the original name to obtain multiple word segmentations, and classify the multiple word segmentations according to preset categories to obtain multiple word segmentations with categories, and then reorganize the multiple word segmentations based on the target category in the categories to obtain multiple candidate names , and perform similarity matching between the multiple candidate names and the preset name database respectively, determine the target candidate name with the highest similarity, and then compare the target candidate name with the name database based on the classification model completed by training. Semantic classification matching, get matching results. In this way, after the exact match of the original name fails, the original name can be disassembled according to the category to obtain multiple word segmentations, and then the multiple segmentations can be reorganized around the specified category in the category to obtain multiple new names, which is easy to operate and reduce The range of word segmentation and combination is reduced, and the workload of matching calculation is reduced. Moreover, for the identified target candidate names, the method of semantic classification and matching is used to convert the traditional similarity matching problem into a machine classification problem, and the similarity of the name pairs is considered from the semantic similarity, which improves the matching accuracy and further reduces the to match the computational workload.

Referring to FIG. 1 , a flow chart of steps of an embodiment of a name matching method of the present invention is shown, which may specifically include the following steps:

Step 101 , in response to a search request for the original name to be searched, matching the original name based on the characters.

Among them, the embodiment of the present invention can be applied to a back-end server, and the back-end server can perform data interaction with the front-end device. The user enters the name to be searched (referred to as "original name") in the front-end device, and initiates a search instruction, and the front-end device receives After the search instruction, the original name can be used to generate a search request, and then the search request can be sent to the backend server. After the backend server receives the search request, it can precisely match the characters of the original name with the preset name database to determine the name database. Is there a name with the same characters as the original name in .

Wherein, the name database includes at least one stored name (referred to as "preset name"); the name may be the name of an enterprise, company, etc., such as "Beijing One Two Three Culture Communication Company".

In this embodiment of the present invention, the character-based matching of the original name includes:

detecting whether a name with the same characters as the original name exists in the name database;

If it exists, the match succeeds; if it does not exist, the match fails.

Specifically, in the case of exact matching, it can detect whether the characters of the original name and any preset name are exactly the same, if there is any preset name that is exactly the same, then the match is successful, if it is not exactly the same as all preset names , then the match fails.

For example, the original name is "Beijing One Two Three Culture Communication Co., Ltd.", assuming that a preset name in the name database is also "Beijing One Two Three Culture Communication Co., Ltd.", then the exact match is successful. The default name is "Beijing One Two Three Culture Communication Company", then the exact match fails.

When the exact match is successful, just return the matching result.

Step 102 , if the matching fails, split the original name to obtain a plurality of word segments, and classify the plurality of word segments according to preset categories to obtain a plurality of word segments with categories.

When the exact match fails, the original name can be split to obtain multiple participles, and then each participle can be classified to determine the one-to-one corresponding category of each participle.

In the embodiment of the present invention, the original name is split to obtain a plurality of word segments, and the plurality of word segments are classified according to preset types to obtain a plurality of word segments with categories, including:

Use jieba to split the original name to obtain multiple participles;

Match each participle with a preset category library, determine the one-to-one corresponding category of each participle, and obtain multiple participles with categories.

Specifically, when splitting the original name, the word segmentation tool jieba can be used to split the original name to obtain multiple word segmentations. Then, each participle is matched with a preset category library to determine the one-to-one corresponding category of each participle.

Among them, the names of enterprises, companies, etc. are generally composed of four parts: region, keyword, industry, and suffix. Since the number of words for region, industry, and suffix is limited, three category libraries can be preset in this embodiment of the present invention, which are respectively : Region library, industry library and suffix library.

Regional libraries include but are not limited to: Beijing, Shanghai, Sichuan, Hebei, Hunan, Shaanxi, Yunnan, Henan, Gansu, Shandong, Hubei, Guangxi, Anhui, Jiangxi, Xinjiang, Shanxi, Fujian, Inner Mongolia, Zhejiang, Heilongjiang, Guizhou, Jilin , Qinghai, Tibet, Liaoning, Guangdong, Jiangsu, Hainan, Ningxia, Shenzhen, Chongqing, Hong Kong, Macau, Taiwan.

Industry libraries include but are not limited to: information, technology, commerce, trade, service, advertising, technology, culture, media, communication, development, communication, consultation, information, management, design, maintenance, logistics, training, design, leasing, construction , engineering, equipment.

The suffix library includes but is not limited to: company, limited company, limited liability company, limited liability company, branch company, limited liability company, and joint stock company.

In this way, after obtaining a plurality of segmented words, each segmented word is matched with the above-mentioned three category libraries respectively, and the one-to-one corresponding category of each segmented word can be determined.

For example, following the example above, the participles of "Beijing One Two Three Culture Communication Company" will be divided into "Beijing", "One Two Three", "Culture", "Communication" and "Company", and then each participle will be divided into Match with each category library, and determine the category of each participle as shown in Table 1:

category value area Beijing Key words one two Three industry culture, communication suffix company

Table 1

It should be noted that the classification method of categories and the values included in the categories may be other than those described above, and may also be other division methods and other values, which may be set according to actual requirements in practical applications. This is not limited.

Step 103: Recombining the multiple word segmentations based on the target category in the category to obtain multiple candidate names.

After obtaining multiple word segmentations with categories, the multiple word segmentations can be reorganized with the target category in the multiple categories as the center to obtain multiple candidate names.

In this embodiment of the present invention, the multiple word segmentations are reorganized based on the target category in the category to obtain multiple candidate names, including:

Recombining the word segments with the target category and the word segments with the non-target category among the plurality of word segmentations to obtain a plurality of candidate names.

Specifically, among the four categories mentioned above, the most important category in the names of enterprises and companies is "keywords". Therefore, each word segmentation can be reorganized with the keywords as the center. The strategy of reorganization includes but not Limited to: Keyword1, Keyword2, …, Keywordn, Keyword1+Keyword2, …, Region+Keyword1, Region+Keyword2,…, Keyword1+Industry,… ….

For example, following the example above, the multiple candidate names obtained after the reorganization of "Beijing One Two Three Culture Communication Company" can be shown in Table 2:

Table 2

Step 104: Perform similarity matching between the plurality of candidate names and a preset name database respectively, and determine the target candidate name with the highest similarity.

After obtaining multiple candidate names, each candidate name can be matched with the name database for similarity, and the similarity corresponding to each candidate name can be obtained, and then the candidate name with the highest similarity can be determined from all the candidate names as Target candidate name.

In the embodiment of the present invention, the similarity matching is performed between the multiple candidate names and the preset name database respectively, and the target candidate name with the highest similarity is determined, including:

Carrying out similarity matching between the multiple candidate names and the name database respectively, and determining the candidate similarity of each candidate name;

The target candidate similarity with the highest similarity among the plurality of candidate similarities is determined, and the candidate name corresponding to the target candidate similarity is used as the target candidate name.

Specifically, for any candidate name among all the candidate names, the similarity can be matched with all the preset names in the name database, so as to obtain the candidate similarity corresponding to each candidate name one-to-one, Then, the target candidate similarity with the highest similarity is determined from the multiple candidate similarities, and the candidate name corresponding to the target candidate similarity is used as the target candidate name.

For example, for the 7 candidate names in Table 2, match each candidate name with the name database, calculate the similarity corresponding to each candidate name one-to-one, and then compare the candidate name with the highest similarity among the 7 similarities. as the target candidate name.

In the embodiment of the present invention, performing similarity matching between the multiple candidate names and the name database, and determining the candidate similarity of each candidate name, includes:

For any candidate name in the plurality of candidate names, perform similarity matching between the any candidate name and at least one preset name in the name database to obtain at least one similarity;

A candidate similarity with the highest similarity among the at least one similarity is determined.

Specifically, for any candidate name among the multiple candidate names, the similarity can be matched with all the preset names in the name database to obtain multiple similarities, and then the similarity with the highest similarity can be matched. degree as the candidate similarity of any candidate name. In this way, the candidate similarity corresponding to each candidate name one-to-one can be obtained.

For example, following the previous example, assuming that the name database includes 50 preset names, for the "one two three" in the seven candidate names, the similarity between "one two three" and the 50 preset names in the name database is matched, 50 similarities are obtained, and then the similarity with the highest similarity among the 50 similarities is used as the candidate similarity. In this way, the candidate similarity corresponding to the seven candidate names one-to-one can be calculated.

Wherein, performing similarity matching between any candidate name and at least one preset name in the name database to obtain at least one similarity, including:

For any preset name in the at least one preset name, obtain the forward maximum common substring and the backward maximum common substring of the any candidate name and the any preset name;

Calculate the forward similarity based on the forward maximum common substring, and calculate the backward similarity using the backward maximum common substring;

Based on the forward similarity and the backward similarity, the similarity between any candidate name and any preset name is calculated.

The shortest edit distance similarity algorithm is generally used when calculating string matching. It considers the semantic relationship between text contexts as a whole. It is a commonly used distance function measurement method and has been widely used in the field of string similarity matching. Applications. The algorithm refers to the minimum number of editing operations required to convert the source string S to the target string T. The fewer operations required, the higher the similarity between the two strings. There are three basic editing operations: ① insert a character in the string S; ② delete a character in the string S; ③ replace a character in the string S with a character in the string T.

In the traditional shortest edit distance algorithm, the similarity between two strings can be calculated from the shortest edit distance. Intuitively, the smaller the edit distance between two strings, the higher the similarity. The formula for converting the edit distance into a similarity with a value in the [0,1] interval is as follows:

where |S|, |T| represent the lengths of strings S and T, respectively, and ld represents the shortest edit distance between strings S and T. The larger the sim(S, T), the higher the similarity between the two strings.

However, the traditional shortest edit distance algorithm only considers the influence of the number of editing operations and the maximum string length, and does not consider the influence of common substrings between strings, so it does not have universal applicability. For example, if the strings are: S ₁ ='BC', S ₂ ='CD', S ₃ ='EF', the similarity between the two strings is calculated according to formula 1 as follows:

1) S ₁ to S ₂ require 1-step replacement and 1-step deletion, and the shortest edit distance is 2, then

0。2) S ₁ to S ₃ require 2-step replacement operations, and the shortest edit distance is 2, then

0.

It can be known from the calculation result that S ₁ and S ₂ have the same similarity as S ₁ and S ₃ , but it is obvious that the similarity of S ₁ and S ₂ is greater than that of S ₁ and S ₃ because the difference between S ₁ and S ₂ The greatest common substring "C" exists. The so-called maximum common substring refers to the string sequence X. If it is a subsequence of two strings, and is the longest among all the sequences that meet this condition, then X is called the maximum common substring of two known sequences.

In order to improve the above-mentioned defects of the shortest edit distance algorithm, the solution of the present invention proposes a shortest edit distance similarity algorithm using the forward maximum common substring and the backward maximum common substring.

Specifically, for any candidate name among the multiple candidate names, denoted as S, and any preset name among all the preset names, denoted as T, the forward maximum common substring of the two is obtained, denoted as is lcs, and the backward maximum common substring, denoted as rcs, and then use formula (2) to calculate the forward similarity

And, using formula (3) to calculate the backward similarity

Among them, |S|, |T| represent the lengths of strings S and T, respectively, |lcs|, |rcs| represent the lengths of the forward maximum common substring lcs and the backward maximum common substring rcs, respectively, and ld represents the string The shortest edit distance between S and T. The forward maximum common substring refers to the largest common substring of the two strings from left to right, and the backward maximum common substring refers to the largest common substring of the two strings from right to left.

Then the forward similarity, backward similarity, and the weight of the two are used to calculate the final similarity, as shown in formula (4):

Among them, α and β are weights, respectively, and α+β=1.

For example, following the example above, assuming α=0.5 and β=0.5, use Equation 4 to calculate the similarity:

It can be seen from the calculation results that the similarity of S ₁ and S ₂ is greater than that of S ₁ and S ₃ , which is more in line with the actual situation than the traditional calculation result of the shortest edit distance.

In the embodiment of the present invention, the target candidate similarity with the highest similarity among the multiple candidate similarities is determined, and the candidate name corresponding to the target candidate similarity is used as the target candidate name, including:

Normalize the multiple candidate similarities based on the forward maximum common substring and the backward maximum common substring to obtain the target candidate similarity with the highest similarity;

The candidate name corresponding to the similarity of the target candidate is used as the target candidate name.

In the application of string similarity matching, when the sequence lengths of the largest common substrings differ greatly, it will have a certain impact on the final similarity calculation result, so it is necessary to normalize the similarity, so as to solve the traditional problem. The calculation defect of the shortest edit distance algorithm greatly improves the matching accuracy. Specifically, after obtaining the candidate similarity corresponding to each candidate name one-to-one, the forward maximum common substring and the backward maximum common substring can be used to normalize the similarity of multiple candidates, as shown in formula (5 ) as shown:

In this way, the final target candidate similarity can be determined from the multiple candidate similarities, and then the candidate name corresponding to the target candidate similarity can be used as the final target candidate name.

Step 105 , perform semantic classification matching between the target candidate name and the name database based on the classification model that has been trained to obtain a matching result.

After the target candidate name is obtained, it is input into the trained classification model, so that the classification model classifies and matches the target candidate name with all preset names in the name database, and obtains a matching result of whether it matches.

In the embodiment of the present invention, the target candidate name is input into a trained classification model, so that the classification model uses a preset feature index to perform semantic classification matching between the target candidate name and the name database;

If the matching is successful, the matching preset name will be used as the matching result; if the matching fails, the matching failure information will be generated.

Specifically, the target candidate name is input into the trained classification model, and the classification model uses preset feature indicators to classify and match the target candidate name and all preset names in the name database. The preset feature indicators include but are not limited to table 3:

table 3

Among them, the customer name is the name of the company recorded and stored in the background server, which is also the name to be matched; the company name is the name of the company obtained by crawling on the web page, that is, the name to be searched.

During classification matching, the classification model determines whether the target candidate name and all preset names represent the same semantics. If the semantics are the same as any preset name, it is determined as 1, that is, the matching is successful, and any preset name is determined as 1. The name is used as the matching result; if it is semantically different from all the preset names, it is judged as 0, that is, the matching fails, and a matching failure message is generated. In this way, the problem of text semantic similarity matching is transformed into a classification problem, which solves the problem of low matching rate caused by only considering literal similarity and ignoring semantic similarity.

Further, classification models can be generated by:

1) Determine the sample set.

Specifically, obtain all customer names in the backend server, as well as all the company names obtained from the web page, and then randomly select a certain number (such as 1%) from the customer names as customer name samples, and also from the company name. A certain number (for example, 1%) is randomly selected as the enterprise name sample, and the customer name sample and the enterprise name sample are used as the final training sample set. Then, the matching name pairs in the training sample set (one matching name pair is "customer name-enterprise name") are set as classification labels. The label is 1 for semantic similarity, and the label is 0 for semantically dissimilar. The remaining customer names and business names are used as a test sample set.

2) Set the characteristic index.

That is, the feature indicators used for classification are set for the classification model, including but not limited to the various feature indicators shown in Table 3.

The feature index includes at least two aspects: on the one hand, calculating various similarity features between name pairs, and on the other hand, calculating NLP (Natural Language Processing, natural language processing) data features of the name pairs.

3) Establish a classification learning model.

The embodiment of the present invention uses a two-layer stacking method to build a machine learning model: the first layer of stacking selects three base classifiers, GussianNBClassifier, RandomForestClassifier, and LogisticRegression as the stacking base model, and the second layer of stacking selects the RandomForestClassifier classifier for training.

Among them, the training model parameters are set as: learning rate ρ=0.001, loss function adjustment factors α=0.25, γ=0.15, which can minimize the model loss function and achieve the optimal solution.

4) Training the classification learning model.

Use the above parameter settings and test sample set to train and verify the classification model, that is, calculate the matching result between the customer name and the reorganized enterprise name, and output the classification label 0 or 1.

Using 1) to 4) to reduce the classification model for many times, the comprehensive evaluation index (F value) of the final classification model can reach 0.79, and the matching accuracy rate of enterprise name and customer name can reach more than 90%, which is efficient and accurate. Business name and customer name matching.

In this embodiment of the present invention, in response to a search request for the original name to be searched, the background server matches the original name based on characters; if the matching fails, the original name is split to obtain multiple word segments, Classify the plurality of word segmentations according to preset categories to obtain a plurality of word segmentations with categories, and then reorganize the plurality of word segmentations based on the target categories in the categories to obtain a plurality of candidate names, and combine the The plurality of candidate names are respectively matched with the preset name database for similarity, and the target candidate name with the highest similarity is determined, and then the target candidate name is semantically classified and matched with the name database based on the classification model completed by training, get matching results. In this way, after the exact match of the original name fails, the original name can be disassembled according to the category to obtain multiple word segmentations, and then the multiple segmentations can be reorganized around the specified category in the category to obtain multiple new names, which is easy to operate and reduce The range of word segmentation and combination is reduced, and the workload of matching calculation is reduced. Moreover, for the identified target candidate names, the method of semantic classification and matching is used to convert the traditional similarity matching problem into a machine classification problem, and the similarity of the name pairs is considered from the semantic similarity, which improves the matching accuracy and further reduces the to match the computational workload.

Further, when dismantling the original name, the similarity algorithm of forward maximum common substring and backward maximum common substring is used, which solves the calculation defect that the similarity of the traditional edit distance only considers the number of edits. Applicability, and greatly improve the matching accuracy.

Further, when determining the target candidate name, the traditional shortest edit distance algorithm ignores the influence of the common length of the string on the edit distance, so there is a certain deviation for the calculation of long string insertion and deletion errors, etc., the present invention implements In this example, an improved normalization method is proposed to solve this problem and further improve the matching accuracy.

It should be noted that, for the sake of simple description, the method embodiments are described as a series of action combinations, but those skilled in the art should know that the embodiments of the present invention are not limited by the described action sequences, because According to embodiments of the present invention, certain steps may be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of the present invention.

Referring to FIG. 2, a structural block diagram of an embodiment of a name matching apparatus of the present invention is shown, which may specifically include the following modules:

a first matching module 201, configured to match the original name based on characters in response to a search request for the original name to be searched;

The word segmentation module 202 is configured to split the original name if the matching fails to obtain a plurality of word segmentations, and classify the plurality of word segmentations according to preset categories to obtain a plurality of word segmentations with categories;

A reorganization module 203, configured to reorganize the multiple word segmentations based on the target category in the category to obtain multiple candidate names;

The second matching module 204 is configured to perform similarity matching between the multiple candidate names and a preset name database respectively, and determine the target candidate name with the highest similarity;

The classification module 205 is configured to perform semantic classification matching between the target candidate name and the name database based on the trained classification model to obtain a matching result.

In the embodiment of the present invention, the first matching module is specifically used for:

detecting whether a name with the same characters as the original name exists in the name database;

If it exists, the match succeeds; if it does not exist, the match fails.

In the embodiment of the present invention, the word segmentation module is specifically used for:

Use jieba to split the original name to obtain multiple participles;

Match each participle with a preset category library, determine the one-to-one corresponding category of each participle, and obtain multiple participles with categories.

In the embodiment of the present invention, the reorganization module is specifically used for:

Recombining the word segments with the target category and the word segments with the non-target category among the plurality of word segmentations to obtain a plurality of candidate names.

In this embodiment of the present invention, the second matching module includes:

a similarity matching submodule, used to perform similarity matching between the multiple candidate names and the name database respectively, and determine the candidate similarity of each candidate name;

The determination submodule is used to determine the target candidate similarity with the highest similarity among the multiple candidate similarities, and use the candidate name corresponding to the target candidate similarity as the target candidate name.

In this embodiment of the present invention, the similarity matching sub-module includes:

a matching unit, configured to perform similarity matching between any candidate name in the plurality of candidate names and at least one preset name in the name database to obtain at least one similarity;

A determining unit, configured to determine a candidate similarity with the highest similarity among the at least one similarity.

In this embodiment of the present invention, the matching unit is specifically used for:

For any preset name in the at least one preset name, obtain the forward maximum common substring and the backward maximum common substring of the any candidate name and the any preset name;

Calculate the forward similarity based on the forward maximum common substring, and calculate the backward similarity using the backward maximum common substring;

Based on the forward similarity and the backward similarity, the similarity between any candidate name and any preset name is calculated.

In this embodiment of the present invention, the determining submodule is specifically used for:

Normalize the multiple candidate similarities based on the forward maximum common substring and the backward maximum common substring to obtain the target candidate similarity with the highest similarity;

The candidate name corresponding to the similarity of the target candidate is used as the target candidate name.

In the embodiment of the present invention, the classification module is specifically used for:

Inputting the target candidate name into the trained classification model, so that the classification model uses a preset feature index to perform semantic classification matching between the target candidate name and the name database;

If the matching is successful, the matching preset name will be used as the matching result; if the matching fails, the matching failure information will be generated.

As for the apparatus embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and reference may be made to the partial description of the method embodiment for related parts.

The embodiment of the present invention also provides an electronic device, including:

It includes a processor, a memory, and a computer program stored on the memory and capable of running on the processor. When the computer program is executed by the processor, each process of the above-mentioned name matching method embodiment can be realized, and the same technology can be achieved. The effect, in order to avoid repetition, is not repeated here.

Embodiments of the present invention further provide a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, each process of the foregoing name matching method embodiment can be achieved, and the same technical effect can be achieved , in order to avoid repetition, it will not be repeated here.

The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments may be referred to each other.

It should be understood by those skilled in the art that the embodiments of the embodiments of the present invention may be provided as a method, an apparatus, or a computer program product. Accordingly, embodiments of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present invention may take the form of a computer program product implemented on one or more computer-usable storage media having computer-usable program code embodied therein, including but not limited to disk storage, CD-ROM, optical storage, and the like.

Embodiments of the present invention are described with reference to flowcharts and/or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the present invention. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing terminal equipment to produce a machine that causes the instructions to be executed by the processor of the computer or other programmable data processing terminal equipment Means are created for implementing the functions specified in the flow or flows of the flowcharts and/or the blocks or blocks of the block diagrams.

These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing terminal equipment to operate in a particular manner, such that the instructions stored in the computer readable memory result in an article of manufacture comprising instruction means, the The instruction means implement the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

These computer program instructions can also be loaded on a computer or other programmable data processing terminal equipment, so that a series of operational steps are performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby executing on the computer or other programmable terminal equipment The instructions executed on the above provide steps for implementing the functions specified in the flowchart or blocks and/or the block or blocks of the block diagrams.

Although preferred embodiments of the embodiments of the present invention have been described, additional changes and modifications to these embodiments may be made by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the embodiments of the present invention.

Finally, it should also be noted that in this document, relational terms such as first and second are used only to distinguish one entity or operation from another, and do not necessarily require or imply these entities or that there is any such actual relationship or sequence between operations. Moreover, the terms "comprising", "comprising" or any other variation thereof are intended to encompass non-exclusive inclusion, such that a process, method, article or terminal device comprising a list of elements includes not only those elements, but also a non-exclusive list of elements. other elements, or also include elements inherent to such a process, method, article or terminal equipment. Without further limitation, an element defined by the phrase "comprises a..." does not preclude the presence of additional identical elements in the process, method, article or terminal device comprising said element.

A name matching method and a name matching device provided by the present invention have been introduced in detail above. The principles and implementations of the present invention are described with specific examples in this paper. The descriptions of the above embodiments are only for help. Understand the method of the present invention and its core idea; at the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation and application scope. In summary, the content of this specification does not It should be understood as a limitation of the present invention.

Claims (12)

1. A method of name matching, the method comprising:

in response to a search request for an original name to be searched, matching the original name based on characters;

if the matching fails, splitting the original name to obtain a plurality of participles, and classifying the participles according to preset categories to obtain a plurality of participles with categories;

recombining the multiple participles based on a target category in the categories to obtain multiple candidate names;

respectively carrying out similarity matching on the candidate names and a preset name database to determine a target candidate name with the highest similarity;

and performing semantic classification matching on the target candidate name and the name database based on the trained classification model to obtain a matching result.

2. The name matching method of claim 1, wherein said matching the original name based on characters comprises:

detecting whether a name identical to the character of the original name exists in the name database;

if yes, matching is successful; if not, the matching fails.

3. The name matching method according to claim 1, wherein the splitting the original name to obtain a plurality of segmented words, and classifying the plurality of segmented words according to a preset type to obtain a plurality of segmented words with categories includes:

splitting the original name by adopting jieba to obtain a plurality of participles;

and matching each participle with a preset category library, determining the category corresponding to each participle one by one, and obtaining a plurality of participles with the categories.

4. The name matching method according to claim 1, wherein the recombining the plurality of participles based on a target category in the categories to obtain a plurality of candidate names comprises:

and recombining the participles with the target category and the participles with the non-target category respectively to obtain a plurality of candidate names.

5. The name matching method according to claim 1, wherein the similarity matching of the candidate names with a preset name database is performed to determine a target candidate name with the highest similarity, and the method includes:

respectively carrying out similarity matching on the candidate names and the name database, and determining the candidate similarity of each candidate name;

and determining the target candidate similarity with the highest similarity in the candidate similarities, and taking the candidate name corresponding to the target candidate similarity as the target candidate name.

6. The name matching method according to claim 5, wherein the similarity matching of the candidate names with the name database and the determination of the candidate similarity of each candidate name comprises:

for any candidate name in the candidate names, performing similarity matching on the candidate name and at least one preset name in the name database to obtain at least one similarity;

and determining the candidate similarity with the highest similarity in the at least one similarity.

7. The name matching method according to claim 6, wherein the similarity matching of any one of the candidate names with at least one preset name in the name database to obtain at least one similarity comprises:

aiming at any preset name in the at least one preset name, acquiring a forward maximum common substring and a backward maximum common substring of the any candidate name and the any preset name;

calculating forward similarity based on the forward maximum common substring, and calculating backward similarity by using the backward maximum common substring;

and calculating the similarity between any candidate name and any preset name based on the forward similarity and the backward similarity.

8. The name matching method according to claim 5, wherein the determining a target candidate similarity with the highest similarity among the plurality of candidate similarities and taking a candidate name corresponding to the target candidate similarity as a target candidate name comprises:

normalizing the candidate similarities based on the forward maximum common substring and the backward maximum common substring to obtain a target candidate similarity with the highest similarity;

and taking the candidate name corresponding to the target candidate similarity as a target candidate name.

9. The name matching method according to claim 1, wherein the performing semantic classification matching on the target candidate name and the name database based on the trained classification model to obtain a matching result comprises:

inputting the target candidate name into a trained classification model, so that the classification model performs semantic classification matching on the target candidate name and the name database by adopting a preset characteristic index;

if the matching is successful, taking the matched preset name as a matching result; and if the matching fails, generating matching failure information.

10. A name matching apparatus, characterized in that the apparatus comprises:

the device comprises a first matching module, a second matching module and a searching module, wherein the first matching module is used for responding to a searching request aiming at an original name to be searched and matching the original name based on characters;

the word segmentation module is used for splitting the original name to obtain a plurality of words if the matching fails, and classifying the plurality of words according to preset categories to obtain a plurality of classified words;

the recombination module is used for recombining the multiple participles based on a target category in the categories to obtain multiple candidate names;

the second matching module is used for respectively carrying out similarity matching on the candidate names and a preset name database to determine a target candidate name with the highest similarity;

and the classification module is used for carrying out semantic classification matching on the target candidate name and the name database based on the trained classification model to obtain a matching result.

11. An electronic device, comprising: processor, memory and computer program stored on the memory and executable on the processor, the computer program, when executed by the processor, implementing the steps of the name matching method as claimed in any one of claims 1 to 9.

12. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the name matching method according to any one of claims 1 to 9.

Images