Name disambiguation method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of name disambiguation, and discloses a name disambiguation method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a to-be-disambiguated name data set, and forming a pairwise-combined name combination according to names; acquiring industry information of each name in the name combination; setting the weight of each preset dimension according to the industry information; calculating a dimension value of the name combination in a preset dimension, and performing weighted summation according to the dimension value of the preset dimension and the weight to obtain a weighted score value; judging whether the weighted score value is greater than a preset weighted score value or not; and affirming the names in the name combinations with the weighted score values greater than a preset weighted score value as the same name. The method has the beneficial effects that the name disambiguation efficiency and accuracy are remarkably improved, the problem of identity confusion of natural persons among different enterprises is effectively solved, and the storage overhead and the calculation complexity are greatly reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of person name disambiguation, and in particular to a person name disambiguation method and device, electronic equipment and a storage medium. BACKGROUND

[0002] It is common for natural persons to serve as directors, supervisors and senior managers, legal persons or investors in different enterprises. Due to regulatory restrictions, enterprise credit license holders cannot obtain the identity card information of natural persons, resulting in identity confusion of natural persons with the same name in cross-enterprise association analysis. The current way of person name disambiguation is mainly through graph neural networks for disambiguation, for example: the public number CN113326377A proposes a multi-strategy person name disambiguation method based on a knowledge graph, which learns node embedding vectors through GCN (graph convolution network). However, this scheme requires maintaining a graph containing ten million nodes and one hundred million associated edges, and the storage overhead increases linearly with the number of nodes, and the time required for a single inference is long. SUMMARY

[0003] Therefore, it is necessary to propose a person name disambiguation method, device, electronic equipment and storage medium for the existing person name disambiguation problem.

[0004] A person name disambiguation method, the method comprising: obtaining a person name data set to be disambiguated, and forming a person name combination combined two by two according to the person name; obtaining industry information of each person name in the person name combination; setting a weight of each preset dimension according to the industry information; calculating a dimension value of the person name combination in the preset dimension, and performing weighted summation according to the dimension value and the weight of the preset dimension to obtain a weighted score value; determining whether the weighted score value is greater than a preset weighted score value; identifying the person name in the person name combination as the same person name when the weighted score value is greater than the preset weighted score value.

[0005] Further, the step of calculating the dimension value of the person name combination in the preset dimension comprises: obtaining a dimension feature of a preset dimension corresponding to two person names in the person name combination; calculating a dimension feature correlation of each preset dimension as a dimension value of the corresponding preset dimension.

[0006] Further, before the step of calculating the dimension feature correlation of each preset dimension as a dimension value of the corresponding preset dimension, the method further comprises: classifying each preset dimension based on a preset dimension feature classification method to obtain a dimension category corresponding to each preset dimension; acquire a standardization processing mode corresponding to the dimension category of each preset dimension; process the corresponding dimension feature based on the standardization processing mode to obtain a dimension feature processed by standardization.

[0007] Further, the dimension category includes three dimension categories of core identity feature, correlation relationship feature and text feature.

[0008] Further, the step of acquiring a person name disambiguation data set and forming a person name combination combined in pairs according to person names comprises: preprocess the person name disambiguation data set to obtain a preprocessed standard person name disambiguation data set; wherein the preprocessing comprises removing the title of each person name in the person name disambiguation data set, deleting abnormal names with less than 2 or more than 5 characters, and merging homophonic characters; count the number of homonyms in the standard person name disambiguation data set; combine the person names in pairs to obtain a person name combination combined in pairs.

[0009] Further, the step of setting the weight of each preset dimension according to the industry information comprises: determine whether the industry information of the two person names in the person name combination belongs to the same industry; If the industry information of the two person names in the person name combination belongs to the same industry, acquire the weight of each preset dimension corresponding to the industry based on a preset correspondence relationship table between the industry and the weight of each preset dimension; wherein the correspondence relationship table is a preset correspondence relationship between the industry and the weight of each preset dimension.

[0010] Further, after the step of determining whether the industry information of the two person names in the person name combination belongs to the same industry, further comprising: If not, acquire the baseline weight of each preset dimension as the weight of each preset dimension; wherein the baseline weight of each preset dimension is a preset weight.

[0011] A person name disambiguation device, the device comprises: A forming module for acquiring a person name disambiguation data set and forming a person name combination combined in pairs according to person names; An acquisition module for acquiring the industry information of each person name in the person name combination; A setting module for setting the weight of each preset dimension according to the industry information; A calculation module for calculating the dimension value of the person name combination in the preset dimension, and performing weighted summation according to the dimension value and weight of the preset dimension to obtain a weighted score value; a judging module configured to judge whether the weighted score value is greater than a preset weighted score value; a judging module configured to judge whether the weighted score value is greater than a preset weighted score value;

[0012] An electronic device comprises a memory and a processor, the memory stores a computer program, and the computer program is executed by the processor to make the processor execute the following steps: obtain a person name data set to be disambiguated, and form a person name combination combined two by two according to the person names; obtain industry information of each person name in the person name combination; set the weight of each preset dimension according to the industry information; calculate the dimension value of the person name combination in the preset dimension, and perform weighted summation according to the dimension value and the weight of the preset dimension to obtain a weighted score value; judge whether the weighted score value is greater than a preset weighted score value; identify the person names in the person name combination whose weighted score value is greater than the preset weighted score value as the same person name.

[0013] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to make the processor execute the following steps: obtain a person name data set to be disambiguated, and form a person name combination combined two by two according to the person names; obtain industry information of each person name in the person name combination; set the weight of each preset dimension according to the industry information; calculate the dimension value of the person name combination in the preset dimension, and perform weighted summation according to the dimension value and the weight of the preset dimension to obtain a weighted score value; judge whether the weighted score value is greater than a preset weighted score value; identify the person names in the person name combination whose weighted score value is greater than the preset weighted score value as the same person name.

[0014] The beneficial effects of the present application are as follows: by constructing two-by-two combinations of person names and dividing the weight in combination with industry characteristics, disambiguation processing can be quickly and accurately performed without relying on large-scale graphs, and the weighted scoring mechanism makes the identification of person name combinations more flexible and accurate, which can effectively distinguish the same name natural persons in a variable enterprise environment. The efficiency and accuracy of person name disambiguation are significantly improved, the problem of identity confusion of natural persons in different enterprises is effectively solved, compared with the current scheme mainly relying on graph neural networks, the present method avoids the need to maintain a large knowledge graph, thereby greatly reducing the storage overhead and computational complexity. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0016] Wherein: Figure 1 An application environment diagram of the person name disambiguation method in an embodiment; Figure 2 A flowchart of the person name disambiguation method in an embodiment; Figure 3 A structural block diagram of the person name disambiguation device in an embodiment; Figure 4 A structural block diagram of the electronic device in an embodiment. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only constitute some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0018] Figure 1 An application environment diagram of the person name disambiguation in an embodiment. Referring to Figure 1 , the person name disambiguation method is applied to a person name disambiguation system. The person name disambiguation system includes a terminal 110 and a server 120. The terminal 110 and the server 120 are connected through a network, and the terminal 110 can be specifically a desktop terminal or a mobile terminal, and the mobile terminal can be specifically at least one of a mobile phone, a tablet computer, a notebook computer, etc. The server 120 can be implemented by an independent server or a server cluster composed of multiple servers. The terminal 110 is used to obtain a person name data set to be disambiguated, and the server 120 is used to disambiguate the person names in the person name data set to be disambiguated.

[0019] As Figure 2 shown, in an embodiment, a person name disambiguation method is provided. The method can be applied to a terminal or a server, and the present embodiment is exemplified by application to a terminal. The person name disambiguation method specifically includes the following steps: S1: obtaining a person name data set to be disambiguated, and forming two-person name combinations according to the person names; S2: obtaining industry information of each person name in the person name combination; S3: Set the weight of each preset dimension according to the industry information; S4: Calculate the dimension value of the person name combination in the preset dimension, and perform weighted summation according to the dimension value and weight of the preset dimension to obtain a weighted score value; S5: Determine whether the weighted score value is greater than a preset weighted score value; S6: Identify the person name in the person name combination whose weighted score value is greater than the preset weighted score value as the same person name.

[0020] As described in the above step S1, obtain the person name data set to be disambiguated, and form two-person name combinations according to the person names. Collect data sets containing person names to be disambiguated, which can come from different sources, such as social media, online forums, articles, books or databases, etc. The person names to be disambiguated usually refer to those that may have the same name, such as "Li Wei" or "Zhang Wei", and the phenomenon of the same name is common in various social and commercial occasions. In order to better handle these same name situations, the person names are combined to form two-person name combinations. For example, if there are two "Li Na" in the data set, the system will combine them into a combination "(Li Na1, Li Na2)". This combination method can help subsequent disambiguation processing, because different spellings, writing habits or related information of the same person may be reflected in multiple person name combinations. It is also possible to merge homophonic characters based on pinyin + character shape similarity (e.g. "Zhang Li" → "Zhang Li").

[0021] As described in the above step S2, obtain the industry information of each person name in the person name combination. After completing the two-person name combination, the second step is to collect the industry information related to these person names. Industry information refers to the professional field, industry background and professional skills associated with each person name. This process can be achieved through data mining technology, web crawler or through existing statistical data. For example, a person named "Li Wei" may work in the IT industry, while another person named "Li Wei" may be a doctor. By accurately collecting these industry information, the system can obtain information from public databases, professional social networking sites (such as LinkedIn) and other channels to distinguish between people with the same name. At the same time, these industry information can provide an important basis for subsequent weighted processing, which helps to select more authoritative or reliable features for judgment in the subsequent steps. Obtaining industry information is a crucial step in the disambiguation process, because different industries may have different effects on the meaning of the person name and its performance in the social network.

[0022] As described in step S3 above, the weight of each preset dimension is set according to the industry information. After the collection of industry information is completed, the next step is to set the weight for the obtained industry information. Specifically, the system will give different weights to the information of each dimension according to different industry characteristics and relevance. For example, some industries such as the technology industry may be given a higher weight due to its high recognition and influence, while some other industries may have less impact on disambiguation and accordingly be given a low weight. The process of setting the weight not only needs to consider the social influence of the industry, but also needs to combine the quality and reliability of the data. This process usually needs to use certain standards or algorithms, such as using expert evaluation, statistical analysis or machine learning models to ensure the scientificity of the weight allocation. The accurate setting in this stage will directly affect the accuracy and effectiveness of the subsequent calculation of weighted score value, so the weight setting needs to be handled carefully in practical application.

[0023] As described in step S4 above, the dimension value of the person name combination in the preset dimension is calculated, and a weighted sum is performed according to the dimension value and weight of the preset dimension to obtain a weighted score value. The value of each person name combination in each preset dimension represents its performance or characteristic degree in that dimension, such as industry experience years, social influence score, etc. Then, the system performs multiplication operation on these dimension values and the previously set weights to calculate the weighted score value of each person name combination. This process is essentially a comprehensive analysis of the information of each dimension, and the contributions of different dimensions are integrated into a single score value through weighted sum. This method can effectively simplify the complex analysis results into a quantitative result, which is convenient for subsequent processing and judgment. The weighted score value obtained finally will become an important basis for judging whether the person names belong to the same person, which can help the system to identify the case where the same name but the industry background and influence are significantly different, thereby reducing misjudgment.

[0024] As described in step S5 above, it is judged whether the weighted score value is greater than a preset weighted score value. After calculating the weighted score value of each combination of person names, it is necessary to make a judgment to determine whether these score values are higher than a preset threshold. By setting a standard, it is determined which person name combination can be considered to refer to the same person. For example, if the set threshold is 80 points, and the weighted score value of a certain combination is 85 points, the system will consider that the person name combination represents the same person. This preset threshold should be reasonably set according to the statistical analysis of historical data and industry standards to ensure the accuracy and reliability of the system. A moderate threshold can help the system to accurately distinguish the case where the same name but actually not the same person during disambiguation, especially in the case where there are a large number of similar person names. Reasonable setting of the threshold is particularly important. Through this judgment, the system can further improve the accuracy of person name disambiguation and reduce the information processing errors caused by confusion.

[0025] As described in step S6, the person name in the person name combination with a weighted score value greater than the preset weighted score value is determined as the same person name. The last step is to determine the person name in the person name combination after judgment as the same person, which is the person name in the combination with a weighted score value greater than the preset threshold value. The implementation of this process needs to classify the person names meeting the conditions into the same category, which is convenient for subsequent data processing and information integration. The decision basis for determining the same person name is the analysis results of multiple dimensions, so as to improve the accuracy of identification, which not only helps to eliminate data redundancy, but also provides more accurate results in subsequent applications such as social recommendation, information recommendation, etc. Determining the same person name may also involve user privacy and data security, so attention should be paid to follow the corresponding legal regulations and ethical norms when implementing. Through this determination, the system can effectively disambiguate the person name and improve the efficiency of information sorting and retrieval.

[0026] In one embodiment, the step S4 of calculating the dimension value of the person name combination in the preset dimension comprises: S401: obtaining the dimension characteristics of the two person names in the person name combination corresponding to the preset dimension; S402: calculating the dimension characteristic correlation of each preset dimension as the dimension value corresponding to the preset dimension.

[0027] As described in step S401, the dimension characteristics of the two person names in the person name combination corresponding to the preset dimension are obtained. The specific dimension characteristics of the two person names in each person name combination in each preset dimension are obtained from the previously collected industry information and feature data. These dimension characteristics may include various information, such as educational background, work experience, social influence, professional skills, and industry awareness, etc. For example, if the person name in the combination is “Zhang Wei”, the information such as his educational institution, work company, position, and popularity in the industry can be obtained. In order to increase the accuracy of the characteristics, various data sources such as professional social platforms, online databases, news reports, etc. can be used to analyze the person name related information in the text through natural language processing technology. In actual operation, the system needs to integrate information from different sources to ensure the completeness and timeliness of the characteristics. In addition, if the system faces incomplete or missing feature information, it may need to use completion technology or machine learning models to predict the missing part in order to obtain complete dimension characteristics. Effective dimension characteristic acquisition can effectively support subsequent dimension correlation calculation, thereby improving the accuracy of person name disambiguation.

[0028] As described in step S402 above, the correlation degree of each preset dimension feature is calculated as the dimension value of the corresponding preset dimension. The correlation between the collected dimension features is calculated to determine their similarity and consistency in each preset dimension. This step is achieved by calculating the strength of different feature relationships, usually using statistical methods or machine learning algorithms such as cosine similarity, Pearson correlation coefficient, etc. These methods can help the system quantify the relationship between two dimension features, resulting in a numerical value as the "dimension value" of that dimension. Specifically, assuming a certain preset dimension, such as "educational background", the system may calculate the correlation of two names in that dimension, such as whether they graduated from the same college, whether they have the same major, the degree of time overlap in their educational experience, etc. Through the comparison of these dimension features, the system can not only determine whether the name combination is similar, but also reveal potential connections and inconsistencies, ensuring the scientific nature of subsequent disambiguation. The accuracy of this process has a direct impact on the final weighted score, so it is important to ensure the reliability and accuracy of the data source when calculating. After calculating the correlation, the system generates a set of numerical values as the basis for subsequent weighted calculation. These values not only reflect the theoretical relationship between features, but also provide operational basis for disambiguation, helping the system to determine which name has consistency in multiple dimensions and ultimately determine whether they belong to the same name. Through this mechanism, the system can effectively distinguish individuals with the same name but different backgrounds, improving the overall disambiguation capability.

[0029] In one embodiment, before the step S402 of calculating the correlation degree of each preset dimension feature as the dimension value of the corresponding preset dimension, the method further comprises: S4021: classifying each preset dimension based on the preset dimension feature classification method to obtain the dimension category corresponding to each preset dimension; S4022: obtaining the standardization processing method corresponding to the dimension category of each preset dimension; S4023: processing the corresponding dimension feature based on the standardization processing method to obtain the standardized dimension feature.

[0030] As described in step S4021 above, each preset dimension is classified based on the preset dimension feature classification method to obtain the dimension category corresponding to each preset dimension. According to the preset dimension feature classification standard, different dimensions are classified. The "preset dimension" here is usually defined based on business needs and data characteristics, and usually involves multiple fields such as education, business, social influence, etc. For example, the dimensions can be classified into personal information dimensions, professional information dimensions, social network dimensions, etc. The purpose of classification is to more accurately analyze and process different types of features. This process usually requires the use of data analysis methods and algorithms, such as clustering analysis, decision trees, etc., to classify similar features into the same category. By classifying the preset dimension features, the system can clearly identify the features that should be focused on under each dimension, thereby simplifying subsequent processing steps. For example, in the education dimension, the focus may be on educational background, graduation institution, major, etc., while in the professional dimension, the focus is on work experience, job history, skills, etc. The output after classification will lay the foundation for subsequent standardization processing and relevance calculation, providing a structured framework for analyzing the similarity of names.

[0031] As described in step S4022 above, the standardization processing method corresponding to the dimension category of each preset dimension is obtained. The method of standardization processing is determined for each dimension category. The purpose of standardization processing is to convert data of different types, ranges, and measures into a comparable format, so that they can be analyzed on the same basis. Since in actual applications, the dimension features of names often come from different databases and information sources, the numerical range, distribution, and type of these features can vary greatly. For example, a certain dimension may be quantitative data (such as work experience), while another may be categorical data (such as awards received).

[0032] Therefore, for each dimension category, the system will define the appropriate standardization processing method. For example, for quantitative data, minimum-maximum standardization (scaling data to the range of [0, 1]) or Z-score standardization (transforming data into a distribution with a mean of 0 and a standard deviation of 1) can be used, while for category data, one-hot encoding and other methods are used to convert it into numerical form. The key to determining the standardization processing method is to ensure the comparability between different dimensions, so that the correlation can be reasonably calculated and weighted summation in the subsequent steps, which provides effective support for disambiguation. Specifically, the core identity feature standardization processing method: contact number: unified international area code (+86), remove special characters (spaces, -, ()), distinguish fixed phone (including area code) and mobile phone, filter invalid numbers such as 11111111, verify the consistency of the registered address and the registered address of the enterprise (inconsistent marked as low weight feature); email address: based on regular expression to verify the format, unify the domain name in lowercase (such as GMAIL.COM→gmail.com), merge well-known email aliases (such as googlemail.com→gmail.com); legal person / steward name: remove titles such as "Mr. / General Manager", merge homophones based on pinyin+character similarity (such as "Zhang Li"→"Zhang Li"), filter abnormal names with length <2 or >5; number of common names: count the number of common names between enterprises, and classify them into 1-5 people (1 person / 2 people / ... / 5 people and above); intersection name score: based on the frequency statistics of rareness of national names (rareness=1 / frequency), normalized to [0, 1].The association relationship feature standardization processing mode: investment relationship path: remove the circular investment path (such as A→B→A), keep ≤3 layers of effective path, and divide according to holding (>50%) / shareholding (20-50%) / small amount of holding (<20%); enterprise group label: generate label based on font similarity (≥80%) + historical name change + investment relationship, and divide into core enterprise / first-level subsidiary / second-level subsidiary; branch relationship: identify based on "branch" / "subsidiary" keywords, and distinguish between general branch and parent-subsidiary company according to the difference in registered capital (subsidiary registered capital ≥ 50% of parent company); legal relationship: identify the cross-employment of the legal persons between enterprises, and record the overlap degree of the legal person's name + tenure; legal supervisor combination: completely match the legal person + supervisor name combination; text feature standardization processing mode: registered address: split the provincial / municipal / county administrative units at three levels, extract the house number / street / building number, and replace synonyms (road = road, hutong = alley); administrative division: provincial / municipal / county three-level matching score (provincial 0.3 + municipal 0.3 + county 0.4); enterprise name: remove regional / organizational form words (such as "Beijing / Co. Ltd"), keep 2-4 core font, and the longer the length, the higher the weight (each increase of 2 words + 0.1); industry word: mapped to the first-level category of "National Economic Industry Classification" (GB / T 4754-2017), and combined "e-commerce / electronic commerce"→"internet retail"; job consistency: predefine the job level weight (legal person = 5, director = 3, supervisor = 2, etc.), and calculate the job similarity of the same name natural person.

[0033] As described in the above step S4023, the corresponding dimension feature is processed based on the standardization processing mode to obtain the standardized dimension feature. The collected dimension features are processed to obtain the standardized dimension features. This process is a direct implementation of applying the standardization method to the previously obtained original dimension feature data. The data after standardization processing can eliminate the influence of dimension and scale between different features, so that the values of each dimension can be compared at the same level. Effective standardization processing can significantly improve the accuracy and robustness of analysis, ensuring that the model will not be affected by the relative influence of other dimensions due to the large numerical range of a certain dimension. Through the standardized dimension features, the system will ultimately reduce confusion and reduce computational complexity, thereby improving the accuracy and efficiency of name disambiguation.

[0034] In one embodiment, the dimension categories include three dimension categories of core identity features, association relationship features, and text features.

[0035] In one embodiment, the step S1 of obtaining the to-be-disambiguated name data set and combining the names in pairs to form name combinations includes: S101: Preprocess the to-be-disambiguated name data set to obtain a preprocessed standard to-be-disambiguated name data set; wherein the preprocessing includes removing the title of each name in the to-be-disambiguated name data set, deleting abnormal names with less than 2 or more than 5 characters, and merging homophonic characters; S102: Count the number of homonyms in the standard to-be-disambiguated name data set. S103: Combine the names with more than one homonym two by two to obtain a two-by-two combined name combination.

[0036] As described in step S101, the to-be-disambiguated name data set is preprocessed to obtain a preprocessed standard to-be-disambiguated name data set. A series of preprocessing is performed on the to-be-disambiguated name data set to obtain a standardized and cleaner data set. Remove the title: The name data set may include titles such as "professor", "doctor", "engineer", etc. These titles have no actual meaning for subsequent disambiguation processing. Specifically, a set of rules can be defined or natural language processing techniques can be used to identify and remove these titles to ensure that the extracted name only includes the core name part. Delete abnormal names: In the data set, some names may not conform to the actual name structure due to the character limit (e.g. only one character or more than five characters). These abnormal names should be deleted. By setting a character limit, invalid or unreasonable names can be screened out. Merge homophonic characters: Homophonic characters refer to names that sound the same but have different forms, such as "Li" and "Li" in spoken language. This may cause confusion. By merging homophonic characters, the complexity of the disambiguation process can be reduced, ensuring that the system can more accurately identify the actual number of homonyms.

[0037] As described in step S102, the number of homonyms in the standard to-be-disambiguated name data set is counted. After the preprocessing of the data is completed, the system may use hash tables or dictionaries and other data structures to efficiently record and update the name statistics when counting the number of homonyms. For example, for each processed name, the system checks the number of times it appears in the statistics table and performs the corresponding accumulation. The advantage of this method is that it can quickly process large amounts of data and complete the statistics within O(n) time complexity. The statistical results can not only be used to identify which names are repeated, but also provide an important basis for subsequent disambiguation analysis. In actual operation, the system selects names with more than one occurrence as potential homonyms, forming the basis data for subsequent combination analysis.

[0038] As described in step S103 above, the same name number greater than 1 is combined with each other to obtain the combination of the same name. After successfully counting the number of the same name, the step S103 is to combine the names with the number of the same name greater than 1 with each other to generate the name combination, which is a very important step in the disambiguation process, aiming to build a combination data for subsequent more in-depth analysis and processing. Specifically, the system will select those names with the number of the same name greater than 1 in the standard disambiguation name data set, and then use nested loop algorithm strategy to generate the combination between each name. For example, if there are three same names “Li Wei” in the data set, and the remaining names are “Zhang Wei” and “Wang Wei”, the system will generate combinations such as (Li Wei, Zhang Wei), (Li Wei, Wang Wei) and (Zhang Wei, Wang Wei). In this way, by combining each pair of the same names, the system can provide potential disambiguation analysis basis for each combination. This process ensures that all possible same name combinations are considered, providing the necessary data basis for subsequent dimension analysis and score calculation. These combinations can not only be used as input data in the subsequent disambiguation process, but also help to identify how different individuals are differentiated in multiple dimensions. Under the implementation of this step, the system not only improves the comprehensiveness of data processing, but also lays the foundation for in-depth analysis of disambiguation.

[0039] In one embodiment, the step S3 of setting the weight of each preset dimension according to the industry information comprises: S301: judging whether the industry information of two names in the name combination belongs to the same industry; S302: if the industry information of two names in the name combination belongs to the same industry, obtaining the weight of each preset dimension corresponding to the industry based on a preset corresponding relationship table between the industry and the weight of each preset dimension; wherein the corresponding relationship table is a preset corresponding relationship between the industry and the weight of each preset dimension.

[0040] As described in step S301, it is determined whether the industry information of the two names in the name combination belongs to the same industry. Because the names in the same industry often have similar professional background, skills and social influence, etc. characteristics, it can better provide correlation basis for disambiguation. The way to execute this judgment may include directly comparing the industry labels of the two names, such as "computer science", "medicine", "education" and the like. In implementation, the system can store the industry information as a string form, a classification label or a numerical index, and determine whether the two industry information are the same through corresponding algorithms (such as direct matching, hash comparison, etc.). This process not only focuses on the consistency of the name of the industry, but also needs to consider the subdivision of the industry. For example, although "computer science" and "information technology" are similar in some ways, their subdivision fields may affect the setting of the weight required for name disambiguation. Therefore, the accuracy of the industry judgment directly affects the scientificity of the subsequent weight setting.

[0041] As described in step S302, if the industry information of the two names in the name combination belongs to the same industry, the weight of each preset dimension corresponding to the industry is obtained based on a preset corresponding relationship table between the industry and the weight of each preset dimension. If the name combination is confirmed to be in the same industry, the weight of each dimension corresponding to the industry is obtained based on a preset corresponding relationship table between the industry and the weight of each preset dimension. The core of this step is to ensure that the system can use the existing industry information to accurately set the weight of each dimension, so as to provide a basis for subsequent score calculation. The preset corresponding relationship table is usually created by domain experts or through data analysis at the beginning of system construction. This table may list multiple industries such as "information technology", "medical", "finance", etc., and assign corresponding weights to different dimensions such as "education background", "work experience", "industry influence" and the like under each industry. For example, for the "information technology" industry, it may be found that the related dimension "work experience" is assigned a higher weight, and the weight of "award experience" is relatively low. The process of obtaining the weight can be realized through simple table lookup or database query. The system will directly retrieve the weight value of the related dimension from the corresponding relationship table for the industry of the current name combination. The accurate implementation of this step will ensure that the result of the subsequent weighted summation has higher relevance and accuracy, so as to more effectively distinguish individuals with the same name but different actual backgrounds. Through the in-depth understanding of the characteristics of the industry, the system can perform more accurately in disambiguation processing, and minimize the need for manual intervention.

[0042] In one embodiment, after the step S301 of determining whether the industry information of the two names in the name combination belongs to the same industry, the method further comprises: S3011: If the industries are not the same, obtaining a baseline weight of each preset dimension as the weight of each preset dimension; wherein the baseline weight of each preset dimension is a pre-set weight.

[0043] As described in step S3011 above, if the industries are not the same, a baseline weight of each preset dimension is obtained as the weight of each preset dimension. That is, the industry information of the two personal names in the personal name combination does not belong to the same industry. In this case, due to the difference in industries, the comparability of the features may be reduced, so the system will adopt a different processing strategy, obtain the baseline weight of each preset dimension, and use it as the weight of each preset dimension, to ensure the correctness and effectiveness of subsequent analysis. First of all, the setting of the baseline weight is usually completed in the system design stage, which is determined through the opinions of industry experts, market research or existing data analysis results. These weights are based on the understanding of the characteristics of different industries and take into account the general applicability of each dimension in the disambiguation process. The baseline weight not only provides a relatively stable perspective, but also allows the system to perform reasonable analysis when more specific industry information is lacking. When performing this step, the system will extract the baseline weight of each dimension from a pre-set weight configuration table. For example, if the preset dimensions include "work experience", "education background" and "industry influence", and assuming that the baseline weight of work experience is 0.5, the baseline weight of education background is 0.3, and the baseline weight of industry influence is 0.2, then in the case where the two industries are different, these pre-set baseline weights will be directly used. In this way, the system can ensure that even in the case of different industries, the dimension features still have a certain reference value, thereby supporting subsequent disambiguation analysis.

[0044] Referring to Figure 3 The present application also provides a personal name disambiguation device, comprising: A forming module 902 is configured to obtain a personal name data set to be disambiguated, and form two-by-two combinations of personal names according to the personal names. An obtaining module 904 is configured to obtain industry information of each personal name in the personal name combination. A setting module 906 is configured to set weights of each preset dimension according to the industry information. A calculating module 908 is configured to calculate dimension values of the personal name combination in the preset dimensions, and perform weighted summation according to the dimension values and weights of the preset dimensions to obtain a weighted score value. A judging module 910 is configured to judge whether the weighted score value is greater than a pre-set weighted score value. An identifying module 912 is configured to identify the personal names in the personal name combination as the same personal name when the weighted score value is greater than the pre-set weighted score value.

[0045] In an embodiment, the calculating module 908 comprises: a dimension feature acquisition submodule, configured to acquire dimension features of two personal names in the personal name combination corresponding to a preset dimension; a dimension value acquisition submodule, configured to calculate a dimension feature correlation degree of each preset dimension as a dimension value corresponding to the preset dimension.

[0046] In an embodiment, the calculating module 908 comprises: a classification submodule, configured to classify each preset dimension based on a preset dimension feature classification manner to obtain a dimension category corresponding to each preset dimension; a standardization processing manner acquisition submodule, configured to acquire a standardization processing manner of each preset dimension corresponding to the dimension category; a dimension feature acquisition submodule, configured to process the dimension feature corresponding thereto based on the standardization processing manner to obtain a dimension feature processed by standardization.

[0047] In an embodiment, the dimension category comprises three dimension categories of core identity feature, correlation relationship feature and text feature.

[0048] In an embodiment, the forming module 902 comprises: a preprocessing submodule, configured to pre-process the to-be-disambiguated personal name data set to obtain a pre-processed standard to-be-disambiguated personal name data set; wherein the pre-processing comprises removing a title of each personal name in the to-be-disambiguated personal name data set, deleting an abnormal name with a word number less than 2 or greater than 5 and merging homophonic characters; a statistical submodule, configured to count a number of homonym in the standard to-be-disambiguated personal name data set; a combining submodule, configured to combine two by two the personal names with a number of homonym greater than 1 to obtain a two-by-two combined personal name combination.

[0049] In an embodiment, the setting module 906 comprises: a judgment submodule, configured to judge whether industry information of two personal names in the personal name combination belongs to a same industry; a corresponding relationship setting submodule, configured to, if the industry information of two personal names in the personal name combination belongs to a same industry, acquire a weight of each preset dimension corresponding to the industry based on a preset corresponding relationship table of industry and weight of each preset dimension; wherein the corresponding relationship table is a preset corresponding relationship of industry and weight of each preset dimension.

[0050] In an embodiment, the setting module 906 comprises: The weight acquisition module is used to acquire the baseline weight of each preset dimension as the weight of each preset dimension if they do not belong to the same industry; wherein the baseline weight of each preset dimension is a pre-set weight.

[0051] Figure 4 An internal structural diagram of an electronic device in one embodiment is shown. This electronic device can specifically be a terminal or a server, and more specifically, a computer device. Figure 4 As shown, the electronic device includes a processor, a memory, and a network interface connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program. When executed by the processor, this computer program enables the processor to implement a name disambiguation method. The internal memory may also store a computer program, which, when executed by the processor, enables the processor to implement a name disambiguation method. Those skilled in the art will understand that… Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device to which the present application is applied. The specific electronic device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0052] In one embodiment, an electronic device is provided, including a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the following steps: Obtain the dataset of names to be disambiguated, and form pairs of names based on the names; Obtain the industry information of each name in the name combination; The weights of each preset dimension are set based on the industry information; Calculate the dimension value of the name combination in a preset dimension, and perform a weighted sum based on the dimension value and weight of the preset dimension to obtain a weighted score value; Determine whether the weighted score value is greater than the preset weighted score value; Names in a combination of names whose weighted score is greater than a preset weighted score are considered to be the same person.

[0053] By constructing two-by-two combinations of human names and dividing weights according to industry characteristics, disambiguation can be quickly and accurately processed without relying on large-scale graphs, and the weighted scoring mechanism makes the identification of human name combinations more flexible and accurate, effectively distinguishing natural persons with the same name in a variable enterprise environment. The efficiency and accuracy of human name disambiguation are significantly improved, effectively solving the problem of identity confusion of natural persons in different enterprises. Compared with the current scheme mainly relying on graph neural networks, this method avoids the need to maintain a large knowledge graph, thereby greatly reducing storage overhead and computational complexity.

[0054] In one embodiment, a computer-readable storage medium is proposed, which stores a computer program, when the computer program is executed by a processor, the processor executes the following steps: Obtain a human name data set to be disambiguated, and form two-by-two combinations of human names according to the human names; Obtain industry information of each human name in the human name combination; Set the weight of each preset dimension according to the industry information; Calculate the dimension value of the human name combination in the preset dimension, and perform weighted summation according to the dimension value and weight of the preset dimension to obtain a weighted score value; Determine whether the weighted score value is greater than a preset weighted score value; Identify the human name in the human name combination as the same human name when the weighted score value is greater than the preset weighted score value.

[0055] By constructing two-by-two combinations of human names and dividing weights according to industry characteristics, disambiguation can be quickly and accurately processed without relying on large-scale graphs, and the weighted scoring mechanism makes the identification of human name combinations more flexible and accurate, effectively distinguishing natural persons with the same name in a variable enterprise environment. The efficiency and accuracy of human name disambiguation are significantly improved, effectively solving the problem of identity confusion of natural persons in different enterprises. Compared with the current scheme mainly relying on graph neural networks, this method avoids the need to maintain a large knowledge graph, thereby greatly reducing storage overhead and computational complexity.

[0056] 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. The program can be stored in a non-volatile computer readable storage medium, and when the program is executed, the processes of the above-mentioned embodiment methods can be included. Any reference to memory, storage, database, or other medium used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0057] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0058] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the protection scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method for person name disambiguation, the method comprising: The method comprises: obtaining a person name data set to be disambiguated, and forming a person name combination combined in pairs according to the person names; obtaining industry information of each person name in the person name combination; setting a weight of each preset dimension according to the industry information; calculating a dimension value of the person name combination in the preset dimension, and performing weighted summation according to the dimension value and the weight of the preset dimension to obtain a weighted score value; judging whether the weighted score value is greater than a preset weighted score value; identifying the person names in the person name combination as the same person name if the weighted score value is greater than the preset weighted score value.

2. The person name disambiguation method of claim 1, wherein, The step of calculating the dimension value of the person name combination in the preset dimension comprises: obtaining dimension features of two person names in the person name combination corresponding to the preset dimension; calculating a dimension feature correlation of each preset dimension as a dimension value corresponding to the preset dimension.

3. The person name disambiguation method of claim 2, wherein, Before the step of calculating the dimension feature correlation of each preset dimension as the dimension value corresponding to the preset dimension, the method further comprises: classifying each preset dimension based on a preset dimension feature classification mode to obtain a dimension category corresponding to each preset dimension; obtaining a standardization processing mode of the dimension category corresponding to each preset dimension; processing the dimension feature corresponding to the dimension category based on the standardization processing mode to obtain a standardized dimension feature.

4. The person name disambiguation method of claim 3, wherein, The dimension category comprises three dimension categories of core identity features, correlation features and text features.

5. The person name disambiguation method of claim 1, wherein, The step of obtaining the person name data set to be disambiguated and forming the person name combination combined in pairs according to the person names comprises: preprocessing the person name data set to be disambiguated to obtain a standard person name data set to be disambiguated after preprocessing; wherein the preprocessing comprises removing a title of each person name in the person name data set to be disambiguated, deleting an abnormal name with a word number less than 2 or greater than 5, and merging homophonic characters; counting a number of same names in the standard person name data set to be disambiguated; combining the person names in pairs if the number of same names is greater than 1 to obtain the person name combination combined in pairs.

6. The person name disambiguation method of claim 1, wherein, The step of setting the weight of each preset dimension according to the industry information comprises: judging whether the industry information of two person names in the person name combination belongs to the same industry; if the industry information of two person names in the person name combination belongs to the same industry, obtaining the weight of each preset dimension corresponding to the industry based on a preset corresponding relationship table of the industry and the weight of each preset dimension; wherein the corresponding relationship table is a preset corresponding relationship of the industry and the weight of each preset dimension.

7. The person name disambiguation method of claim 6, wherein, After the step of judging whether the industry information of two person names in the person name combination belongs to the same industry, the method further comprises: if not, obtaining a reference weight of each preset dimension as the weight of each preset dimension; wherein the reference weight of each preset dimension is a preset weight.

8. A person name disambiguation apparatus characterized by comprising: The device comprises: a forming module configured to obtain a person name data set to be disambiguated, and form a person name combination combined in pairs according to the person names; an obtaining module configured to obtain industry information of each person name in the person name combination; a setting module configured to set a weight of each preset dimension according to the industry information; The computing module is configured to calculate dimension values of the person name combination in preset dimensions, and perform weighted summation according to the dimension values of the preset dimensions and weights to obtain a weighted score value. The judging module is configured to judge whether the weighted score value is greater than a preset weighted score value. The identifying module is configured to identify a person name in the person name combination whose weighted score value is greater than the preset weighted score value as the same person name.

9. A computer-readable storage medium, characterized in that, The computer program is stored in the memory and is executed by the processor to enable the processor to perform the steps of the person name disambiguation method according to any one of claims 1 to 7.

10. An electronic device, comprising: The device comprises a memory and a processor, and the memory stores a computer program which is executed by the processor to enable the processor to perform the steps of the person name disambiguation method according to any one of claims 1 to 7.

Images