A character portrait generation method for talent selection

Abstract

The invention discloses a method for generating a character portrait for talent selection, comprising the following steps: S1, managing the ability words, constructing a multi-level corpus system consisting of ability root words, ability derived words, and ability adjacent words; S2, constructing an ability tree model; S3, using the ability tree in S2 and the ability words in S1 to generate an individual's ability matrix from the employee's evaluation document; S4, analyzing the ability matrix, matching qualitative words that can characterize the ability words, and inserting the qualitative words into the ability matrix; S5, statistically analyzing the employee's ability matrix and quantitatively calculating the qualitative results, and finally generating a character portrait. The invention has the beneficial effects that the invention adopts a series of methods based on natural language text processing, abstracts the ability of a character through evaluation documents, and makes the evaluation dimension concrete and numerical.

Description

Technical Field

[0001] The present invention belongs to the field of natural language processing and data mining, is oriented to a personnel system, and particularly relates to a character portrait generation method oriented to talent selection. Background Art

[0002] In the actual business scenarios of the HR department, there are often a large number of evaluation documents for individuals. These documents are difficult to use in the management of HR in a suitable way. There are two main problems: first, there is a lack of a system that can extract character capabilities and character portraits. Second, when there are many employees, the number of related documents will also be very large. It is almost impossible to use evaluation documents manually, which consumes a lot of time and manpower, and lacks quantifiable indicators and persuasiveness. The existing character portrait systems and methods are more oriented to the recommendation of e-commerce business and advertising business, while there are few character portrait methods for the HR field and individual capabilities, and the effect is not good. Summary of the invention

[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for generating character portraits for talent selection by using a series of methods based on natural language text processing to abstract the abilities of characters through evaluation documents, thereby making the evaluation dimensions concrete and numerical.

[0004] The objective of the present invention is achieved through the following technical solution: A method for generating a character portrait for talent selection, comprising the following steps:

[0005] S1. Manage the ability words and build a multi-level corpus system consisting of ability root words, ability words, and ability adjacent words;

[0006] S2, build a capability tree model;

[0007] S3, using the competency tree in S2 and the competency words in S1, generate an individual competency matrix from the employee’s evaluation document;

[0008] S4, analyzing the competence matrix, matching qualitative words that can characterize the competence words, and inserting the qualitative words into the competence matrix;

[0009] S5. Calculate the employee competency matrix and quantitatively calculate the qualitative results to ultimately generate a character portrait.

[0010] Furthermore, the specific implementation method of step S1 is:

[0011] S11. Use natural language tools to count the frequency and select several words with the highest frequency as the ability root words W. root ;

[0012] S12, W rootWith W tail Do the Cartesian product, the result is M candidate , from M candidate Select several ability words W ability Add ability word database T w ;

[0013] S13. Use the skip-gram model and CBOW model in Google's open source word2vec tool to map each ability word W using the evaluation documents as the corpus. ability is a word vector V in a multidimensional space ability =(1, 0, 0, 1, ...);

[0014] S14, setting bias parameter δ;

[0015] S15. Define two word vectors V a and V b The similarity is Sim, which is calculated using the improved Pierce correlation coefficient method. The formula is as follows:

[0016]

[0017] Among them, Va i , Vb i is the word vector V a , the i-th component of Vb, is the word vector V a , the component mean of Vb; n is the dimension of the word vector;

[0018] Set the sim threshold to 0.8, that is, if the similarity value of two word vectors is greater than 0.8, the words are considered to have similar meanings and are added to the adjacent word library T s Medium; T s Each adjacent word W sim T w A capability word W ability Phase mapping.

[0019] Furthermore, the specific implementation method of step S2 is:

[0020] S21. Each node in the capability tree is a TreeNode i {}, the information contained in the node includes: the node number Node_id, the ability word number Word_id, the weight value weight, and the child node list tree_son[]. The complete ability tree node structure is:

[0021] TreeNode i {Node id , Word id,weight,tree_son[]};

[0022] S22. Create a new ability tree AbilityTree i , add the name and description information of the capability tree; then select the name and description information from the existing vocabulary T w and T s Select a capability word as the root node word of the tree and create a new capability node TreeNode i , add the ability word number Word_id to TreeNode i Then fill in the weight value weight and the child node list tree_son[];

[0023] The parent node is TreeNode f Add a new sub-capability node TreeNode son , and fill in the basic information of the child capability node: Node_id of this node, capability word Word_id, weight value weight, child node list tree_son[]; save the child node in the tree_son[] list of the parent node, and the storage structure is as follows: TreeNode f {tree_son[Node_id son ]}.

[0024] Furthermore, the specific implementation method of step S3 is:

[0025] S31. Input the employee number for which the competency matrix is ​​to be generated, retrieve all evaluation documents for the employee, and use the regular expression Z1 to divide the evaluation document into different short sentences S i ; Regular expression Z1 uses punctuation marks comma, period, and space to segment short sentences at the beginning or end:

[0026] Z1=[^. ,]*? {}[^. ]*? [. ,];

[0027] S32, through the ability word database T w and adjacent word database T s Each W ability , W sim , retrieve all the documents of the employee with the phrase S i , save the data containing W ability , W sim The short sentence is S r , as the basis for the next step of characterization of the ability word;

[0028] S33, combine the short sentence with the ability word or adjacent word into a list L i <W|S|W attr >, where W attrIt is the qualitative indicator word of the ability word, and the default value is used by default;

[0029] S34. Matrix the employee's capabilities emp_id It is defined as the set of all lists, and the calculation formula is as follows:

[0030]

[0031] Furthermore, the specific implementation method of step S4 is:

[0032] S41. Input employee ID emp_id and check whether the employee has capability matrix D emp_id , if there is a competency matrix, the matrix is ​​characterized by competency terms; otherwise, the process returns to step S3 to generate the competency matrix of the employee;

[0033] S42, using the part-of-speech tagging method in the natural language processing jieba tool library, extract D emp_id Sentences in i , S i Segment the words and mark the parts of speech one by one, converting them into two character lists Char w and Char p , Char w is the word segmentation result list, Char p The result list of word segmentation and part of speech is consistent in length, ensuring that the index values ​​of vocabulary and part of speech are the same and maintain a 1-to-1 relationship;

[0034] S43, according to the expression of Chinese and evaluation documents, the commonly used expressions are abstracted into part-of-speech characters, and regular matching mode is used to identify S i The converted part-of-speech list Char p , so as to accurately match the ability word W i Qualitative qualitative word W attr , the result W attr Insert into the employee's competency matrix D emp_id middle.

[0035] Furthermore, the specific implementation method of step S5 is as follows:

[0036] S51, construct evaluation mapping table P, P i =(W attr , f score ), f score is the evaluation score, with a value range of [-1, 1]. The evaluation mapping table represents the one-to-one correspondence between evaluation words and scores;

[0037] S52. Traverse employee matrix D emp_id , through the employee matrix W attrQuery the score f in the evaluation mapping table P score , calculate the average of multiple evaluation scores as the final score of the ability; summarize D emp_id Then output the character portrait of the employee's ability.

[0038] The beneficial effects of the present invention are as follows: the present invention adopts a series of methods based on natural language text processing to abstract the capabilities of characters through evaluation documents, so that the evaluation dimensions are concrete and numerical. The character portrait system based on the variable capability model can realize character portraits from unstructured documents, so that in personnel business scenarios, there are more reference bases and another evaluation dimension, which can well solve the dilemma and difficulty of personnel selection in large companies and units, effectively enhance the reliability of personnel decision-making, and greatly improve the efficiency and workflow of personnel screening. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The present invention is a flow chart of a method for generating a character portrait for talent selection. DETAILED DESCRIPTION

[0040] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0041] like Figure 1 As shown, a method for generating a character portrait for talent selection according to the present invention comprises the following steps:

[0042] S1. To create a character portrait from personnel evaluation documents, it is necessary to provide support for the rich and available ability words. First, the ability words should be managed, including the addition, deletion, modification and query of ability words. In the face of the synonym problem in Chinese, the synonym dictionary can be used to generate adjacent words. In the subsequent calculations, the adjacent words are incorporated into the corresponding ability words for statistics. The construction of ability words is the basis for the generation of this character portrait. First, a detailed and reasonable ability word library is constructed; a multi-level corpus system consisting of ability root words, ability words, and ability adjacent words is constructed, so that the content of the corpus can be continuously enriched and the corpus can be expanded from different angles; the specific implementation method is:

[0043] S11. Use natural language tools to count the frequency and select several words with the highest frequency as the ability root words W. root , such as: W root =["responsibility", "accountability", "management"]; Set the ability word suffix to W tail , for example: W tail =[“awareness”, “quality”, “ability”], etc.;

[0044] S12, W root With W tail Do the Cartesian product, the result is M eandidate , from Mcandidate Select several ability words W ability Add ability word database T w ;

[0045]

[0046] S13. Use the skip-gram model and CBOW model in Google's open source word2vec tool to map each ability word W using the evaluation documents as the corpus. ability is a word vector V in a multidimensional space ability =(1, 0, 0, 1, ...);

[0047] S14, in order to avoid similarity calculation accuracy errors and inaccurate similarity, set the bias parameter 6, and manually adjust the value of δ so that the calculation results of the adjacent words tend to be stable and meet the meaning of Chinese synonyms;

[0048] S15. In view of the synonym phenomenon in the evaluation documents, it is necessary to expand the ability vocabulary with adjacent words, classify them as the same ability in subsequent calculations, and improve the reliability and accuracy of the character portrait generation method; use Tencent's open source ChineseEmbedding synonym vocabulary to define two word vectors V a and V b The similarity is Sim, which is calculated using the improved Pierce correlation coefficient method. The formula is as follows:

[0049]

[0050] Among them, Va i , Vb i is the word vector V a , the i-th component of Vb, is the word vector V a , the component mean of Vb; n is the dimension of the word vector;

[0051] Set the sim threshold to 0.8, that is, if the similarity value of two word vectors is greater than 0.8, the words are considered to have similar meanings and are added to the adjacent word library T s Medium; T s Each adjacent word W sim T w A capability word W ability Phase mapping is convenient for subsequent calculation and processing.

[0052] S2. Construct a capability tree model according to the hierarchical relationship of "abstract-concrete" for capability words; for example, the capability words "professional capability" and "professional skills" and "professional quality". "Professional capability" has a higher degree of abstraction and is the root node of the capability tree. "Professional skills" and "professional quality" are the child nodes of the next level. The specific implementation method is:

[0053] S21. Through the "capability tree" model, the expression of capabilities can be freely customized and the general scope of capability words is clearer. The capability tree structure is similar to the tree structure in the data structure; each node in the capability tree is a TreeNode i {}, the information contained in the node includes: the node number Node_id, the ability word number Word_id, the weight value weight, and the child node list tree_son[]. The complete ability tree node structure is:

[0054] TreeNode i {Node id , Word id ,weight,tree_son[]};

[0055] S22. Create a new ability tree AbilityTree i , add the name and description information of the capability tree; then select the name and description information from the existing vocabulary T w and T s Select a capability word as the root node word of the tree and create a new capability node TreeNode i , add the ability word number Word_id to TreeNode i Then fill in the weight value weight and the child node list tree_son[];

[0056] The parent node is TreeNode f Add a new sub-capability node TreeNode son , and fill in the basic information of the child capability node: Node_id of this node, capability word Word_id, weight value weight, child node list tree_son[]; save the child node in the tree_son[] list of the parent node, and the storage structure is as follows: TreeNode f {tree_son[Node_id son ]}.

[0057] S3, using the competency tree in S2 and the competency words in S1, generates a personal competency matrix from the employee's evaluation document; the specific implementation method is:

[0058] S31. Input the employee number for which the competency matrix is ​​to be generated, retrieve all evaluation documents for the employee, and use the regular expression Z1 to divide the evaluation document into different short sentences S i ; Regular expression Z1 uses punctuation marks comma, period, and space to segment short sentences at the beginning or end:

[0059] Z1=[^. ,]*? {}[^. ]*? [. ,];

[0060] S32, through the ability word database T w and adjacent word database T s Each W ability , W sim , retrieve all the documents of the employee with the phrase S i , save the data containing W ability , W sim The short sentence is S r , as the basis for the next step of characterization of the ability word;

[0061] S33, combine the short sentence with the ability word or adjacent word into a list L i <W|S|W attr >, where W attr It is the qualitative indicator word of the ability word, and the default value is used by default;

[0062] S34. Matrix the employee's capabilities emp_id It is defined as the set of all lists, and the calculation formula is as follows:

[0063]

[0064] S4. Analyze the capability matrix, match qualitative words that can characterize the capability words, and insert the qualitative words into the capability matrix; the specific implementation method is:

[0065] S41. Input employee ID emp_id and check whether the employee has capability matrix D emp_id , if there is a competency matrix, the matrix is ​​characterized by competency terms; otherwise, the process returns to step S3 to generate the competency matrix of the employee;

[0066] S42, using the part-of-speech tagging method in the natural language processing jieba tool library, extract D emp_id Sentences in i , S i Segment the words and mark the parts of speech one by one, converting them into two character lists Char w and Char p , Char w is the word segmentation result list, Char pThe result list of word segmentation and part of speech is consistent in length, ensuring that the index values ​​of vocabulary and part of speech are the same and maintain a 1-to-1 relationship;

[0067] During the specific implementation process, the suggestions of experts from the human resources department can be integrated to add or modify regular matching rules.

[0068] S43. Based on the expression of Chinese and evaluation documents, we seek the opinions and suggestions of human resources experts, abstract the commonly used expressions into part-of-speech characters, and use regular matching mode to identify S i The converted part-of-speech list Char p , so as to accurately match the ability word W i Qualitative qualitative word W attr , the result W attr Insert into the employee's competency matrix D emp_id middle.

[0069] The regular matching mode used in this embodiment is shown in Table 1:

[0070] Table 1

[0071]

[0072] S5. Count the employee capability matrix and quantitatively calculate the qualitative results to finally generate a character portrait. The specific implementation method is:

[0073] S51, construct evaluation mapping table P, P i =(W attr , f score ), f score is the evaluation score, and its value range is [-1, 1]. The evaluation mapping table represents the one-to-one correspondence between evaluation words and scores. The evaluation scores in this embodiment are shown in Table 2.

[0074] Table 2

[0075] <![CDATA[W attr ]]> <![CDATA[f score ]]> powerful 0.8 better 0.5 Slightly insufficient -0.2 Very bad -0.8

[0076] S52. Traverse employee matrix D emp_id , through the employee matrix W attr Query the score f in the evaluation mapping table P score , calculate the average of multiple evaluation scores as the final score of the ability; traverse D emp_id During the process, if W sim Belong to T w , then the qualitative score of the adjacent word is added to the corresponding W iability In; Summary D emp_id Then output the character portrait of the employee's ability.

[0077] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for generating a character portrait for talent selection, characterized in that: The following steps are involved: S1. Manage the ability words and build a multi-level corpus system consisting of ability root words, ability words, and ability adjacent words. The specific implementation method is: S11. Use natural language tools to count the frequency and select several words with the highest frequency as the ability root words W. root ; S12, W root With W tail Do the Cartesian product, the result is M candidate , from M candidate Select several ability words W ability Add ability word database T w ; S13. Use the skip-gram model and CBOW model in Google's open source word2vec tool to map each ability word W using the evaluation documents as the corpus. ability is a word vector V in a multidimensional space ability =(1,0,0,1…); S14, setting bias parameter δ; S15. Define two word vectors V a and V b The similarity is Sim, which is calculated using the improved Pierce correlation coefficient method. The formula is as follows: Among them, Va i , Vb i is the word vector V a , the i-th component of Vb, is the word vector V a , the component mean of Vb; n is the dimension of the word vector; Set the sim threshold to 0.8, that is, if the similarity value of two word vectors is greater than 0.8, the words are considered to have similar meanings and are added to the adjacent word library T s middle; T s Each adjacent word W sim T w A capability word W ability Phase mapping; S2. Construct a capability tree model; the specific implementation method is: S21. Each node in the capability tree is a TreeNode i {}, the information contained in the node includes: the node number Node_id, the ability word number Word_id, the weight value weight, and the child node list tree_son[]. The complete ability tree node structure is: TreeNode i {Node id ,Word id ,weight,tree_son[]}; S22. Create a new ability tree AbilityTree i , add the name and description information of the capability tree; then select the name and description information from the existing vocabulary T w and T s Select a capability word as the root node word of the tree and create a new capability node TreeNode i , add the ability word number Word_id to TreeNode i Then fill in the weight value weight and the child node list tree_son[]; The parent node is TreeNode f Add a new sub-capability node TreeNode son , and fill in the basic information of the child capability node: Node_id of this node, capability word Word_id, weight value weight, child node list tree_son[]; save the child node in the tree_son[] list of the parent node, and the storage structure is as follows: TreeNode f {tree_son[Node_id son ]}; S3, using the competency tree in S2 and the competency words in S1, generates a personal competency matrix from the employee's evaluation document; the specific implementation method is: S31. Input the employee number for which the competency matrix is ​​to be generated, retrieve all evaluation documents for the employee, and use the regular expression Z1 to divide the evaluation document into different short sentences S i ; Regular expression Z1 uses punctuation marks comma, period, and space to segment short sentences at the beginning or end: Z1＝[^。，]*？{}[^。]*？[。，]； S32, through the ability word database T w and adjacent word database T s Each W ability , W sim , retrieve all the documents of the employee with the phrase S i , save the data containing W ability , W sim The short sentence is S r , as the basis for the next step of characterization of the ability word; S33, combine the short sentence with the ability word or adjacent word into a list L i <W|S|W attr >, where W attr It is the qualitative indicator word of the ability word, and the default value is used by default; S34. Matrix the employee's capabilities emp_id It is defined as the set of all lists, and the calculation formula is as follows: S4. Analyze the capability matrix, match qualitative words that can characterize the capability words, and insert the qualitative words into the capability matrix; the specific implementation method is: S41. Input employee ID emp_id and check whether the employee has capability matrix D emp_id , if there is an ability matrix, the matrix is ​​characterized by ability words; Otherwise, return to step S3 to generate the employee's competence matrix; S42, using the part-of-speech tagging method in the natural language processing jieba tool library, extract D emp_id Sentences in i , S i Segment the words and mark the parts of speech one by one, converting them into two character lists Char w and Char p , Char w is the word segmentation result list, Char p The result list of word segmentation and part of speech is consistent in length, ensuring that the index values ​​of vocabulary and part of speech are the same and maintaining a 1-to-1 relationship; S43, according to the expression of Chinese and evaluation documents, the commonly used expressions are abstracted into part-of-speech characters, and regular matching mode is used to identify S i The converted part-of-speech list Char p , so as to accurately match the ability word W i Qualitative qualitative word W attr , the result W attr Insert into the employee's competency matrix D emp_id middle; S5. Count the employee capability matrix and quantitatively calculate the qualitative results to finally generate a character portrait. The specific implementation method is: S51, construct evaluation mapping table P, P i =(W attr ,f score ), f score is the evaluation score, with a value range of [-1,1]. The evaluation mapping table represents the one-to-one correspondence between evaluation words and scores; S52. Traverse employee matrix D emp_id , through the employee matrix W attr Query the score f in the evaluation mapping table P score , calculate the average of multiple evaluation scores as the final score of the ability; summarize D emp_id Then output the character portrait of the employee's ability.

Images