CN114612062A

CN114612062A - False recruitment early warning method and system

Info

Publication number: CN114612062A
Application number: CN202210222605.XA
Authority: CN
Inventors: 吴丹
Original assignee: Qianjin Network Information Technology (shanghai) Co ltd
Current assignee: Qianjin Network Information Technology (shanghai) Co ltd
Priority date: 2022-03-09
Filing date: 2022-03-09
Publication date: 2022-06-10

Abstract

The invention relates to a false recruitment early warning method and a system, wherein the method comprises the following steps: extracting target recruitment enterprise information, target position information and interview information from recruitment interview information provided by a user; respectively crawling information in a network according to a plurality of indexes in an enterprise level, a job level and an interview level based on the extracted target recruitment enterprise information, target job information and interview information, and processing the crawled information into corresponding index data; according to the risk assessment strategy, performing risk assessment from an enterprise level, a job level and an interview level according to the index data of each level; and in response to the evaluated risk, backtracking and analyzing the index data corresponding to the risk to obtain early warning information and providing the early warning information for the user. The invention identifies the false recruitment from multiple aspects and multiple dimensions and warns the user in time to avoid the user being cheated.

Description

False recruitment early warning method and system

Technical Field

The invention relates to the technical field of data processing, in particular to a method and a system for early warning of false recruitment.

Background

The false recruitment generally refers to fraudulent recruitment in which recruitment information published on the network and in a talent market does not conform to actual employment situations, and can refer to both false recruitment performed by a regular company and illegal recruitment performed by a false company. Some regular companies may nominal positions and publish recruitment information in the network and/or talent market for reasons such as collecting talent information, promoting corporate awareness, etc. The false recruitment behavior can not only delay the time for job hunting of job seekers and harm personal information safety, but also occupy and waste a job hunting public platform and damage the public credibility of the recruitment platform. Illegal recruitment performed by a false company can cause property and personal damage to job seekers, such as charge disordering, marketing, yellow finance related and the like, and can also damage interests of third parties, such as a qualified company being impersonated and a recruitment platform for issuing recruitment information.

To identify such false recruitment information, the industry has made a great deal of effort to do so. The Chinese patent application with the publication number of CN113704409A and the invention name of 'a false recruitment information detection method based on cascade forests' provides a false recruitment information detection method, and a model is established by using position data issued by a network recruitment platform to predict false recruitment based on a cascade forest algorithm of a decision tree. The Chinese patent application with the publication number of CN113506084A and the invention name of 'a false recruitment position detection method based on deep learning' provides a false recruitment position detection method, collects false recruitment information from an online recruitment platform or a recruitment APP, processes the false recruitment information into sample data of a training model, obtains the detection model through training, and utilizes the detection model to detect the false recruitment information on the online recruitment platform or the recruitment APP.

According to the technical scheme, at present, when the virtual recruitment information is identified, the training data for training the identification model is from the virtual position information and/or the virtual recruitment information collected from the network recruitment platform. The source range of the data is narrow, and the identification dimension is single. According to the previous description of the false recruitment, the false recruitment with the fraudulent nature has multiple aspects, and if the recognition is only limited from the aspect of position, a great amount of the false recruitment is missed. Therefore, a scheme for identifying the false recruitment in multiple aspects and dimensions is needed.

Disclosure of Invention

Aiming at the technical problems in the prior art, the invention provides a false recruitment early warning method and a system, which are used for identifying false recruitment from multiple aspects and multiple dimensions and warning a user in time so as to avoid the user being cheated.

In order to solve the technical problem, according to an aspect of the present invention, there is provided a method for warning of a false recruitment, comprising the steps of:

extracting target recruitment enterprise information, target position information and interview information from recruitment interview information provided by a user;

respectively crawling information in a network according to a plurality of indexes in an enterprise level, a job level and an interview level based on the extracted target recruitment enterprise information, target job information and interview information, and processing the crawled information into corresponding index data;

according to the risk assessment strategy, performing risk assessment from an enterprise level, a job level and an interview level according to the index data of each level; and

and in response to the evaluated risk, backtracking and analyzing the index data corresponding to the risk to obtain early warning information and providing the early warning information for the user.

According to one aspect of the invention, the invention provides a false recruitment early warning system, which comprises a user information acquisition module, a data collection module, an index data generation module, a risk assessment module and an early warning module, wherein the user information acquisition module is configured to receive recruitment interview information provided by a user and extract target recruitment enterprise information, target position information and interview information from the user recruitment interview information; the data collection module is connected with the internet, is connected with the user information acquisition module, and is configured to perform information crawling in the network according to a plurality of indexes in an enterprise level, a job level and an interview level respectively based on the extracted target recruitment enterprise information, target job position information and interview information; the index data generation module is connected with the data collection module and is used for processing the crawled information of the enterprise level, the job level and the interview level to obtain a plurality of index data of corresponding levels; the risk assessment module is connected with the index data generation module and is configured to perform risk assessment from an enterprise level, a position level and an interview level according to the index data of each level according to a risk assessment strategy; the early warning module is connected with the risk assessment module and is configured to respond to the assessed risk, backtrack and analyze index data corresponding to the risk to obtain early warning information and provide the early warning information for the user.

According to the method and the system provided by the invention, the invention makes up the defects that the predecessors cannot actively search information and judge risks: the traditional algorithm only adopts the position description related information to judge whether the recruitment is false, and the invention actively searches other related information, such as enterprise information, related enterprise information, target position information, related position information and the like, thereby increasing the accuracy of judgment. The invention makes up the deficiency that the prior person can not predict the risk in the interview notification link: the existing detection scheme for the false recruitment is earlier than an interview notification link, and the false recruitment which shows the risk only in an interview notification stage cannot be effectively detected, but the method provided by the invention analyzes and evaluates according to interview information given by job seekers after the job seekers obtain interview notifications of target enterprises, so that the time is the last line of defense before the job seekers go to interview, and the best time for predicting the risk by obtaining the most sufficient data can be obtained, and the job seekers can be helped to directionally avoid the risk to the greatest extent. The invention also makes up the deficiency that the predecessor can not accurately position and identify the risk source: based on the relevant data of the interview, the problem of backtracking the whole enterprise level and the job level can be searched upwards after the risk is evaluated, the essential reasons of the risk are combed, the recruitment fraud problem overall picture possibly existing behind the risk is obtained, and a more general evaluation conclusion and a more detailed analysis report are obtained.

Drawings

Preferred embodiments of the present invention will be described in further detail below with reference to the accompanying drawings, in which:

fig. 1 is a flow diagram of a method for false recruitment identification according to one embodiment of the invention;

FIG. 2 is a flow diagram of risk assessment according to one embodiment of the present invention;

FIG. 3 is a flow diagram of risk assessment according to another embodiment of the present invention;

FIG. 4 is a flow chart for performing risk assessment according to yet another embodiment of the present invention;

FIG. 5 is a flow chart for assessing risk according to another embodiment of the present invention;

FIG. 6 is a flow diagram of a method of processing test sample data according to one embodiment of the invention;

fig. 7 is a flow diagram of a false recruitment warning method according to one embodiment of the invention;

fig. 8 is a functional block diagram of a false recruitment identification system provided in accordance with an embodiment of the present invention;

fig. 9 is a partial functional block diagram of a false recruitment recognition system provided in accordance with an embodiment of the present invention;

fig. 10 is a functional block diagram of a portion of a false recruitment identification system according to one embodiment of the present invention;

FIG. 11 is a functional block diagram of a subscriber information acquisition module according to one embodiment of the present invention;

FIG. 12 is a functional block diagram of a data processing system according to one embodiment of the present invention;

FIG. 13 is a functional block diagram of a data collection module according to one embodiment of the present invention;

FIG. 14 is a functional block diagram of an index data generation module according to one embodiment of the present invention;

FIG. 15 is a functional block diagram of a model training module according to one embodiment of the present invention;

fig. 16 is a functional block diagram of a false recruitment warning system according to one embodiment of the present invention;

FIG. 17 is a functional block diagram of an early warning module according to one embodiment of the present invention;

FIG. 18 is a functional block diagram of an early warning module according to another embodiment of the present invention; and

fig. 19 is a functional block diagram of an early warning module according to yet another embodiment of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In the following detailed description, reference is made to the accompanying drawings that form a part hereof and in which is shown by way of illustration specific embodiments of the application. In the drawings, like numerals describe substantially similar components throughout the different views. Various specific embodiments of the present application are described in sufficient detail below to enable those skilled in the art to practice the teachings of the present application. It is to be understood that other embodiments may be utilized and structural, logical or electrical changes may be made to the embodiments of the present application.

Fig. 1 is a flow diagram of a method for false recruitment identification according to one embodiment of the invention. In this embodiment, the method comprises the steps of:

and step S1a, determining whether or not the recruitment interview information provided by the user is received, executing step S2a if the recruitment interview information provided by the user is received, and repeating the step if the recruitment interview information is not received. In one embodiment, when the method is applied to an online recruitment platform, such as a recruitment website, a recruitment APP, etc., the online recruitment platform opens an interface to the user for facilitating the user to input recruitment interview information through the interface. The interface may be implemented, for example, as a user interface through which a user enters recruitment interview information. In some embodiments, the user interface is further provided with a plurality of information sections for inputting the target recruitment enterprise information, the target position information and the interview information. The target recruitment enterprise information comprises information such as an enterprise name and an address. The target job information includes information such as job title, job site, and the like. The interview information includes information such as interview location, interview time, interview notification mode, interview form, such as oral test, written test, etc.

And step S2a, extracting the target recruitment enterprise information, the target position information and the interview information from the recruitment interview information provided by the user. In one embodiment, when a user inputs various information through the information segment in the user interface, the information in the field is read from the information segment, and the information is processed to obtain one or more keywords. Such as deactivating words, word segmentation, etc.

And S3a, based on the extracted target recruitment enterprise information, target position information and interview information, respectively crawling information in the network according to a plurality of indexes in the enterprise level, the position level and the interview level.

Wherein, in one embodiment, the enterprise-level indicators include static indicators including, but not limited to, one or more of the number of enterprise people targeted for recruitment of the enterprise, the number of affiliates, registered funds, financing information, enterprise age, annual business volume, business segment, and enterprise type, and dynamic indicators. The dynamic indicators include, but are not limited to, one or more of enterprise lawsuit events for the targeted recruiting enterprise, major investments accepted, social news events occurring, and social media enterprise ratings. The job level indexes include but are not limited to one or more indexes of job names, job sites, work departments, monthly salary ranges, job descriptions, welfare treatments, job types, experience requirements, academic requirements, school types, job industry risks and repeated recruitment times of the same job. The interview level index includes but is not limited to one or more of interview place, interview time, interview notification mode, interview conducting times, whether a written interview is available and interview form. And acquiring corresponding information from the network according to the indexes, for example, searching the network by taking the name and/or address of the information such as the name and address of the target recruitment enterprise as a keyword according to the information such as the name and address of the target recruitment enterprise provided by the user to obtain the official website, the social media public number, the published news report and the like of the target recruitment enterprise.

In one embodiment, the target recruitment enterprise official network searches and obtains information on the number of enterprises, information including "xx branches", information on the description of business scope, information including categories of enterprises, information on financing, information on the years of enterprises, information on annual business amount, information on registered funds, and the like, and if some information is not obtained from the target recruitment enterprise official network, the search from the network is continued. For example, if the official website of the target recruitment enterprise does not obtain the registered fund, the enterprise type and other information, the registered information of the target recruitment enterprise can be obtained from the industrial and commercial information website, and the registered fund and the enterprise type can be obtained from the registered information. For another example, if the financing information is not obtained in the official website of the target recruitment enterprise, the financing related information can be obtained by inquiring the social media public number or published news report of the target recruitment enterprise. And identifying comment information, litigation information, news information about received investment and the like of the target recruitment enterprises in the last five years on the webpage. In order to obtain the index data of the related position level, position information of the same position and other positions released by the target recruitment enterprise is searched in the plurality of recruitment platforms, and position information of the same position and other positions released by the same type of enterprise as the target recruitment enterprise is searched. And acquiring information related to the position from a target recruitment enterprise official network. And acquiring the post information released by the target recruitment enterprise within a certain time period from the network.

In step S4a, the crawled information is processed into index data of a corresponding level. And processing the crawled corresponding information according to a processing strategy of a specific index, and standardizing the corresponding information into specific index data according to various indexes, wherein the index data can be floating point numbers, character string vectors or certain codes.

For example, the processing of static metrics for enterprise levels includes: and regarding the number of the enterprises, processing the crawled data of the number of the target recruitment enterprises into floating point numerical values. Regarding the number of branches, the number is calculated according to the words containing ". about.. the branches" recognized from the introduction of the official website company, and processed into floating point numerical values; regarding the registered funds, the numerical values are processed into floating point numerical values according to the words containing "registered funds x" identified from the introduction of the company registration information. With regard to scope, keywords in the scope description are identified and processed into a string vector. Regarding the type of the enterprise, the enterprise type information on the official website is processed into a character string vector, and matching is performed in "national enterprise", "foreign fund", "joint fund" and "private camp", and the above options are respectively processed into 3, 2, 1 and 0 and recorded. And regarding financing information, processing the net value in the obtained recent financial report of the enterprise into a floating point numerical value. And regarding the enterprise age, subtracting the acquired enterprise standing time from the current year to obtain a standing year number value, and processing the standing year number value into a floating point number value. Regarding the annual amount, the annual amount is processed as a floating point value based on the annual amount obtained from the latest financial report of the enterprise.

The processing of dynamic metrics for enterprise levels includes: regarding legal litigation events of enterprises, firstly extracting litigation information of a target recruitment enterprise in a recent period (such as five years) from the acquired information; then, performing regular matching on each piece of information by using keywords such as 'labor service', 'being called' and 'target recruitment enterprise name', and if matching is successful, marking as 1, and if matching is failed, marking as 0; when the matching is successful, acquiring litigation time information from the litigation information and encoding, for example, encoding to 1 when the litigation time is less than 1 year; the code is 0.8 when the litigation time is less than 2 years and more than 1 year; the code is 0.6 when the litigation time is less than 3 years and more than 2 years; the code is 0.5 when the litigation time is less than 4 years and more than 3 years; the code is 0.3 when the litigation time is less than 5 years and more than 4 years; multiplying all successfully matched codes by the codes of litigation time to obtain litigation codes; and finally, adding all litigation codes to obtain the litigation event score of the enterprise.

Regarding the major investment accepted by the enterprise, news information about the latest period (such as five years) and the received investment is extracted from the captured information; then processing the received investment amount into a floating point numerical value; adding all the floating point numerical values to obtain the major investment sum accepted by the target recruitment enterprise in the last period (such as five years); if not, the default value is 0.

Regarding social media business evaluation, first extracting a plurality of pieces of comment information of a recent period of time (such as five years) from the captured information; then, each piece of comment information is processed according to the following processing procedures: removing HTML symbols in the comments by an analysis method; removing punctuation marks through a regular expression; screening and removing all stop words by using a stop word library; labeling the cleaned data text; analyzing the text by using an emotion analysis model, and outputting each comment as a comment text emotion code of-1 (negative) or 1 (positive); then obtaining comment time information, wherein the code is 1 when the comment time distance is less than 1 year currently, the code is 0.8 when the comment time distance is less than 2 years currently and more than 1 year currently, and the code is 0.6 when the comment time distance is less than 3 years currently and more than 2 years currently; the code is 0.5 when the comment time distance is currently less than 4 years and greater than 3 years; the code is 0.3 when the comment time distance is currently less than 5 years and greater than 4 years; multiplying all comment text emotion codes by time codes to obtain an emotion score of each comment, for example, if a certain enterprise receives a negative comment and the current time distance is less than 4 years and more than 3 years, the emotion score of the comment is-1 × 0.5 ═ 0.5; the enterprise also receives a positive comment, and the appearing time distance is less than 3 years and more than 2 years, so that the emotion score of the comment is 1 x 0.6-0.6; and adding the emotion scores of all the comments of the enterprise to obtain the social media evaluation score of the enterprise, wherein the social media evaluation score of the enterprise in the example is-0.5 + 0.6-0.1.

The processing of the index data of the job level comprises the following steps: processing the job names into character string vectors; for a work place, firstly processing the extracted work place information into a character string vector; finding corresponding geographical position information through a map search engine; then acquiring the name of an owner/tenant of the position building, and matching the name with the name of the enterprise; and if the target recruitment enterprise name can be matched, the code is 0, and otherwise, the code is 1. For the work department, firstly processing the extracted work department information into a character string vector; then matching the character string with the information on the official website of the company, wherein if the corresponding character string can be matched on the official website, the code is 0, and otherwise, the code is 1. For the monthly salary range, processing the monthly salary range into a floating point numerical value; for job description, removing HTML symbols, spaces and punctuation marks by an analytic method and a regular expression; then identifying and deleting stop words through a stop word library; and finally processing the residual text into a character string vector. Regarding welfare treatment, firstly removing HTML symbols, spaces and punctuation marks through an analytic method and a regular expression; identifying and deleting stop words through a stop word library; and finally processing the residual text into a character string vector. Regarding the work type, whether characters such as 'full time', 'part time', 'labor dispatch' and 'salary' exist in the crawled related information is searched, if the characters exist, the characters are marked as 3, if the characters exist, the characters are marked as 3, the characters exist, the part time of the functions, the characters exist, the characters, the states of the labor dispatch, the states, the existence, the exist, the characters, the exist, the characters, the existence, the characters, the exist, the characters, the exist, the states, the exist, the states, the results, the exist, the states, the existence, the labor dispatch, the states, the existence. Regarding experience requirements, firstly, identifying whether an 'experience' character string is contained in the crawled position related information, if so, identifying a numerical value before the 'year' character string to obtain a numerical value requiring the number of working years, processing the numerical value into a floating point numerical value, and taking the minimum value of the numerical value to record; if not, whether the character string is 'due generation' or not is identified, if yes, the character string is recorded as 0, and if not, the character string is recorded as null. Regarding the academic requirement, the description of the academic requirement is firstly identified from the crawled position related information, and the description is matched in the character strings of 'master or above', 'college subject', 'university subject', 'secondary or below' and 'no requirement', and the options are respectively processed into 4, 3, 2, 1 and 0 for recording. For the school type, descriptions about graduation schools are identified from the crawled position related information, matching is carried out in character strings such as '985' and '211', if '985' is identified, the description is recorded as 2, if '211' is identified, the description is recorded as 1, if '985' and '211' are also recorded as 1, and if no matching field is identified, the description is recorded as 0. For the risk of the work industry, matching the crawled position description information in a classification dictionary of the industry of the people's republic of China to obtain the industry to which the position description information belongs; and then searching related news by taking the affiliated industry as a keyword, if the keywords such as 'referee', 'transformation', 'market evaporation', 'store closing' and the like are searched, defining the keywords as high risk and marking the keywords as 1, and otherwise defining the keywords as low risk and marking the keywords as 0. Regarding the repeated recruitment times of the same post, firstly capturing all released work posts from the crawled target recruitment enterprises in the last period of time (such as three years); then identifying the duration of the same post published on the network and converting the duration into a floating point numerical value according to the year; and if the post is off-line and then on-line again, adding the on-line time lengths respectively and converting the on-line time lengths into floating point numerical values.

The position name, the work place, the work department, the monthly salary range, the position description, the welfare treatment, the work type, the experience requirement, the academic requirement, the school type, the work industry risk, the repeated recruitment times of the same position and the like of the target position are depth indexes of the target position, and a collection after the specification processing of the depth indexes can be used as a prediction sample set of the target position. In another embodiment, the method further comprises the step of establishing an enterprise network recruitment information matrix according to the position-level information while obtaining the target position depth index data, and obtaining related indexes of the recruitment platform and the target position in the breadth, hereinafter referred to as breadth indexes, from the enterprise network recruitment information matrix.

The enterprise network recruitment information matrix comprises one or more pieces of position information released by a target enterprise and recruitment platform qualification information for releasing the position information; the job information released by one or more similar enterprises which are similar to the target enterprise and the qualification information of the recruitment platform for releasing the job information. The positions released by the target enterprise comprise a target position and a second position different from the target position; the recruitment platforms include one or more target recruitment platforms that publish a target position provided by the target enterprise and one or more second recruitment platforms that publish second position information different from the target position.

The step of obtaining the corresponding breadth index data based on the enterprise network recruitment information matrix processing comprises the following steps:

calculating a qualification grade vector of each recruitment platform based on the qualification information of the recruitment platform;

counting the number of positions released by each recruitment platform from the enterprise network recruitment information matrix;

determining a degree of engagement for each recruitment platform based on the number of positions; for example, the number of positions is used as the degree of insertion, or the number of positions is normalized and the normalized value is used as the degree of insertion.

And calculating the qualification grade weighting vector value of each recruitment platform by taking the input degree of each recruitment platform as the qualification grade vector weight of the recruitment platform. Wherein, the qualification grade weighting vector value of the recruitment platform is an extent index.

The step of obtaining the corresponding breadth index data based on the enterprise network recruitment information matrix processing further comprises the following steps:

querying a plurality of second target positions, which are provided by the same type of enterprises and are the same with the target positions, based on the target positions;

calculating a position vector of the target position and each second target position based on the position information;

performing clustering operation on the plurality of second target positions to obtain a second target position with the largest clustering result;

and calculating a first vector difference value of the position vector of the target position and the position vector of a second target position with the largest clustering result, and taking the first vector difference value as an external consistency coefficient of the target position. Wherein the external consistency coefficient of the target position is another breadth index.

and calculating a second vector difference value of the position vector of the target position and the position vector of each second position provided by the target enterprise, and calculating the average value of the second vector difference values, wherein the average value of the second vector difference values is used as an internal consistency coefficient of the target position. Wherein the internal consistency coefficient of the target position is another breadth index.

and determining respective weights according to the affinity and sparseness of the first vector difference value and the second vector difference value average value and the target position vector, and calculating the weighted average value of the first vector difference value and the second vector difference value as a consistency coefficient of the target position. Wherein the consistency coefficient of the target position is another breadth index.

In one embodiment, the depth index data and the breadth index data obtained as described above are merged together as a prediction sample set of the target position.

The processing of the index data of the interview level comprises the following steps: regarding interview sites, firstly, processing interview site information into a character string vector; then finding out corresponding geographical position information through a map search engine; acquiring the name of an owner/tenant of the position building, and matching the name with the name of the target recruitment enterprise; and if the target recruitment enterprise name is matched, the code is 0, and otherwise, the code is 1. For interview time, if the interview time information is between 8:00 and 18:00, it is recorded as 0, otherwise it is recorded as 1. For the interview notification mode, the code is 1 if the interview notification mode is a telephone contact mode, the code is 2 if the interview notification mode is a short message mode, the code is 3 if the interview notification mode is an email mode, the code is 4 if the interview notification mode is a recruitment application (recruitment App) mode, the code is 5 if the interview notification mode is a common social software mode, and a plurality of contact modes are allowed to be filled. Whether a written test exists or not is coded as 0 if the written test exists, and is coded as 1 if the written test does not exist.

And S5a, evaluating according to the index data of each level to obtain enterprise-level risks, position-level risks and interview-level risks of the target enterprise for the false recruitment. In this embodiment, the trained models are used for evaluation when evaluating risks at different levels. The model can adopt algorithms such as a decision tree, naive Bayes, multi-layer, k-nearest neighbor, random forest, neural network and the like.

In this embodiment, the machine learning models at the respective levels are obtained by respectively training labeled samples at the respective levels. The process of training the machine learning model is described with the training of the enterprise-level machine learning model as an example.

The training data of the enterprise-level machine learning model comprises a certain amount of enterprise sample data, each piece of enterprise sample data comprises a plurality of index data marked as a false recruitment enterprise or a real recruitment enterprise, and each index data is used as a feature data of the enterprise sample. For example, the number of enterprises, the number of affiliates, the registered funds, the business scope, the type of enterprise, the financing information, the age of enterprise, the annual business amount, the legal action event of enterprise, the major investment accepted, the social media enterprise rating, etc. are used as arguments of one sample (i.e., one-dimensional characteristics of one sample), and whether there is an enterprise-level risk is manually marked on each sample, for example, a sample with risk is marked as "negative"/"positive", and a sample without risk is marked as "positive"/"negative", so as to obtain a sample set. The sample set should meet some requirements of the training model, such as the number balance of positive and negative samples, the feature dimension of each sample being the same, and so on. And taking 80% of samples in the sample set as training samples and 20% of samples in the sample set as verification samples, thereby respectively forming a training set and a verification set.

And then modeling according to any one algorithm of the algorithms such as the decision tree, naive Bayes, multi-layer, k-nearest neighbor, random forest, neural network and the like. And the model is used for carrying out supervised learning on the samples based on the training set according to a model algorithm to obtain the probability of enterprise-level risks of each sample.

Since the present embodiment only sets two result types, "risky" and "no risk", the model employs a two-class output. Assuming that the number of sample arguments (i.e., the dimensions of the sample features) is n, the mathematical representation of the model is:

X＝{x₁，x₂，…，x_n}

Y＝{y₀，y₁}

wherein X is a feature set of a sample input to the model, each X_iIs a one-dimensional feature. For example: x is the number of₁Number of enterprises, x₂Number of divisions, x₃To register funds, … …, x_nIs the annual business volume. Y is a classified set of models, Y_iE 0,1, in this example, y₀Representing "no risk", y₁Representing "at risk". The model being based on a given sample X_iCalculate it as label c_j(labels for each category y, e.g. y)₀Value 0, corresponding label c₀Is "no enterprise level risk") probability p (c)_j|X_i) Then the output of the model is:

wherein

Representing the class prediction, p (c)_j|X_i) Representing the probability of each class for a given sample, a contrast threshold is set, in one embodiment, such as 0.6-0.9. In this embodiment, the threshold value of 0.8 is used for explanation, and when judging whether there is enterprise-level risk, if y is output by the algorithm model₁If the prediction probability of (2) exceeds 0.8, the enterprise level risk of the enterprise corresponding to the sample can be confirmed. The threshold value may be an empirical value obtained by repeatedly operating the model according to the practical data, or may be further adjusted along with the iteration of the model. And verifying the trained model by adopting a verification set sample, and when the model meets the evaluation standard, using the model for online risk evaluation.

The training process of the job level machine learning model and the interview level machine learning model is similar to the process, and is not repeated.

In step S4a, in order to make the normalized data conform to the input requirement of the model, a sample is composed of multidimensional features, and the number of features is the same. After the index data of each level are specifically normalized into floating point numbers, character string vectors or encoded, the data are combined into prediction samples of the models according to the input requirements of the models of each level, and then the prediction samples of each level are input into the corresponding models in step S5a, so that the risks corresponding to each level are obtained.

In the foregoing embodiment, the output of each level of machine learning model is "at risk" or "no risk", and in one embodiment, as shown in fig. 2, various prediction samples are input to the machine learning models of the corresponding levels in steps S510a, 511a and 512a, respectively, and then the results output by the machine learning models of each level, namely, enterprise level risk, job level risk and interview level risk, are combined as an evaluation result in step S513a, and the evaluation result is provided to the user in step S6a, so that the user can know whether there is risk from three aspects of enterprise, job and interview.

In another embodiment, as shown in FIG. 3, the "at risk" or "no risk" output by the model is taken as two levels, represented by 1 and 0, respectively, so that the risk levels at the enterprise level, the job level, and the interview level can be encoded to obtain the risk code. After inputting various prediction samples to the machine learning models of the corresponding levels to obtain the risks of the corresponding levels at steps S510a, 511a, and 512a, the enterprise-level risks, job-level risks, and interview-level risks are encoded at step S523 a. For example, risk code 001 represents risk only at interview, while risk code 111 represents risk at business, job, and interview. In order to make the user feel the magnitude of the risk, different risk levels are set in the present embodiment, and the risk codes correspond to the risk levels, as shown in table 1 below:

table 1:

risk coding	Risk rating
			111	High risk
110、101	Higher risk
		100	Middle risk
010、011	Lower risk
		000、001	Low risk

In step S524a, the corresponding table of risk codes and risk levels is queried according to the currently obtained risk codes, as shown in table 1, the risk level corresponding to the currently obtained risk code can be obtained and determined as the final risk level, and then the final risk level is provided to the user as the evaluation result in step S6 a.

In another embodiment, as shown in fig. 4, after the coding of the obtained risk levels of the enterprise level, the job level and the interview level in step S533a, the method further includes:

step S534a, the weights of the enterprise level, the job level, and the interview level are obtained.

Step S535a, a weighted sum of the risk codes is obtained by performing weighted calculation according to the numerical values of the respective bits in the risk codes and the respective weights.

In step S536a, the risk level matching the weighted sum of the current risk codes is obtained by querying the corresponding table of the weighted sum of the risk codes and the risk level, as shown in table 2.

In one embodiment, the weighted sum of the risk codes 000 and 111 is calculated to obtain 0,1, 4, 5, 6, 9, and 10 according to the weights of the enterprise level, the job level, and the interview level being 5, 4, and 1, respectively, so that the following 8 values are divided into 3 groups according to the difference between adjacent values, and the values are (0, 1), (4, 5, 6), (9, and 10), respectively, to obtain the corresponding table 2:

table 2:

for example, in one embodiment, when the following risks are derived from each level model: enterprise level: "at risk"; job level: "No risk"; the face test level: when there is no risk, the risk code is obtained as 100, the weighted sum of the corresponding risk codes is 5, and the table 2 is queried, so that the risk grade corresponding to the risk code is matched as 'medium risk'.

Finally, in step S6a, the risk level queried from table 2 is provided to the user as an evaluation result.

In the above embodiment, the models of the respective levels use two risk levels, i.e., "at risk" and "no risk", but it is of course also possible to train the models of each level to use multi-classification output, for example, when using 5 levels of "high risk", "higher risk", "medium risk", "lower risk" and "low risk", assuming that the number of sample independent variables (i.e., the dimensions of the sample features) is n, the mathematical expression of the model is:

X＝{x₁，x₂，…，x_n}

Y＝{y₀，y₁，y₂，y₃，y₄}

wherein X is a feature set of a sample input to the model, each X_iIs a one-dimensional feature. E.g. x₁Number of enterprises, x₂Number of divisions, x₃To register funds, …, x_nIs the annual business volume. Y is a classified set of models, Y_iE {0,1,2,3,4}, in this example, y₀Representing "low risk", y₁Representing "lower risk", y₂Represents "intermediate risk", y₃Stands for "higher risk", y₄Representing a "high risk". The model being based on a given sample X_iCalculate it as label c_j(labels for each category y, e.g. y)₀Value 0, corresponding label c₀Is "no risk") of probability p (c)_j|X_i) Then the output of the model is:

wherein

Representing the class prediction, p (c)_j|x_i) Representing the probability of each class for a given sample, a contrast threshold is set in one embodiment, typically 0.6-0.9. In the embodiment, the threshold value is 0.8, for example, when judging the enterprise-level risk category, if y outputted by the algorithm model₁If the prediction probability of the sample exceeds 0.8, the enterprise corresponding to the sample can be confirmed to have lower enterprise-level risk. The threshold value may be an empirical value obtained by repeatedly operating the model according to the practical data, or may be further adjusted along with the iteration of the model. And verifying the trained model by adopting a verification set sample, and when the model meets the evaluation standard, using the model for online risk evaluation.

After the risk level of each level is obtained, the final risk level is determined by encoding or calculating the weighted sum of the encoding according to the foregoing embodiment, which is not described herein again.

FIG. 5 is a flow chart of assessing risk according to another embodiment of the present invention. In this embodiment, there is one enterprise-level machine learning model, a plurality of job-level machine learning models and interview-level machine learning models, and two or more output categories of the enterprise-level machine learning model, the job-level machine learning model and the interview-level machine learning model, and the number of the output categories of the three models may be the same or different. The levels are set from top to bottom in sequence according to the sequence from the enterprise level, the position level to the interview level. The number of the next-stage machine learning models is the same as the number of the output categories of the previous-stage machine learning models, and the next-stage machine learning models and the previous-stage machine learning models are trained by data with risks of the output categories of the previous-stage machine learning models respectively. For example, the plurality of job-level machine learning models are respectively trained from data having corresponding enterprise-level risk levels, and respectively correspond to the risk levels output by the corresponding enterprise-level machine learning models; the interview level machine learning models are respectively obtained by training data with corresponding enterprise level risk levels and corresponding job level risk levels and respectively correspond to the risk levels output by the corresponding job level machine learning models.

For example, in one embodiment, the output of the enterprise-level machine learning model and the job-level machine learning model are classified outputs defined as two risk levels, "at risk" and "no risk", respectively. The interview level machine learning model defines 5 levels of "high risk", "higher risk", "medium risk", "low risk", and "no risk", respectively, for example, for multi-classification output. The enterprise-level machine learning models are marked as M1, the number of the position-level machine learning models is two, and the position-level machine learning models are respectively marked as a model M2.1 and a model M2.2, wherein position samples of a training set of the model M2.1 are all samples corresponding to enterprise-level no risk, and position samples of a training set of the model M2.2 are all samples corresponding to enterprise-level risk. The interview level machine learning models are 4 and are respectively recorded as models M3.1, M3.2, M3.3 and model M3.4, the position samples of the training set of the model M3.1 are samples corresponding to enterprise-level no risk and position-level no risk, the position samples of the training set of the model M3.2 are samples corresponding to enterprise-level no risk and position-level risk, the position samples of the training set of the model M3.3 are samples corresponding to enterprise-level risk and position-level no risk, and the position samples of the training set of the model M3.4 are samples corresponding to enterprise-level risk and position-level risk.

According to the training sample of the machine learning model, the enterprise level machine learning model is the first level, the position level machine learning model is the subordinate model of the enterprise level machine learning model, and the interview level machine learning model is the subordinate model of the position level machine learning model. The selection of the lower model should correspond to the output of the higher model, such as risk.

And selecting a next-level machine learning model according to the risk level output by the previous-level machine learning model from top to bottom according to the sequence of the enterprise level, the job level and the interview level, wherein the risk and the level output by the interview-level machine learning model are the final risk and the level thereof. The specific risk assessment process is shown in fig. 5, and includes the following steps:

in step S51a, the enterprise-level prediction samples are input to the enterprise-level machine learning model M1 for prediction. In one embodiment, the enterprise level machine learning model is denoted as M1, and enterprise level prediction samples are input to the model M1.

Step S52a, determine if the output of the enterprise-level machine learning model M1 is "at risk", if "at risk", perform step S53a, and if "no risk", perform step S57 a.

Step S53a, selects the job-level machine learning model M2.2, and inputs the job-level prediction samples to the job-level machine learning model M2.2 for prediction.

Step S54a, determine whether the output of the position-level machine learning model M2.2 is "at risk", if "at risk", execute step S55a, and if "no risk", execute step S56 a.

Step S55a, selecting the interview-level machine learning model M3.4, inputting the interview-level prediction samples into the interview-level machine learning model M3.4 for prediction, taking the output of the interview-level prediction sample input model M3.4 as a risk assessment result, and then ending the risk assessment process.

Step S56a, selecting the interview-level machine learning model M3.3, inputting the interview-level prediction sample to the interview-level machine learning model M3.3 for prediction, taking the output of the interview-level machine learning model M3.3 as a risk assessment result, and then ending the risk assessment process.

Step S57a, the job-level machine learning model M2.1 is selected, and the job-level prediction samples are input to the job-level machine learning model M2.1 for prediction.

Step S58a, determining whether the output of the position-level machine learning model M2.1 is "at risk", if the output of the position-level machine learning model M2.1 is "at risk", executing step S59a, if the output of the position-level machine learning model M2.1 is "no risk", executing step S510a,

step S59a, selecting the interview-level machine learning model M3.2, inputting the interview-level prediction sample to the interview-level machine learning model M3.2 for prediction, taking the output of the interview-level machine learning model M3.2 as a risk assessment result, and then ending the risk assessment process.

Step S510a, selecting the interview-level machine learning model M3.1, inputting the interview-level prediction samples into the interview-level machine learning model M3.1 for prediction, taking the output of the interview-level machine learning model M3.1 as a risk assessment result, and then ending the risk assessment process.

According to the embodiment, all information is divided according to the whole enterprise level, the position level and the interview level, the position level evaluation result is collected by the enterprise level evaluation result, and the interview level evaluation result is collected by the position level evaluation result, so that comprehensive and accurate risk judgment is made.

In this embodiment, the interview-level machine learning model is a multi-class output, and in one embodiment, the interview-level machine learning model obtains each classification probability through calculation of input prediction samples, for example, the probabilities of "high risk", "higher risk", "medium risk", "low risk", and "no risk" are obtained respectively; comparing each classification probability with a corresponding second classification threshold; and if one of the classification probabilities is greater than or equal to the corresponding second classification threshold, determining the risk level defined by the model output for the classification.

In another embodiment, the output of the interview-level machine learning model is in two classes; two risk levels, namely "at risk" and "no risk", are defined respectively; in computing predicted samples, the interview-level machine learning model computes a "at risk" probability; comparing the 'at-risk' probability with a plurality of third classification thresholds, wherein the plurality of third classification thresholds form a plurality of low-to-high risk probability intervals which respectively correspond to a plurality of low-to-high risk levels; and determining the corresponding risk level according to the risk probability interval where the 'at risk' probability output by the interview level machine learning model is located.

In another embodiment, the output of the enterprise-level machine learning model and the job-level machine learning model may also be output as interview-level machine learning models, i.e., as a multi-classification output. Each job-level machine learning model corresponds to a class of outputs of the enterprise-level machine learning model, and thus the number of job-level machine learning models is the same as the number of classes of risk outputs of the enterprise-level machine learning model. Correspondingly, the number of interview level machine learning models is the product of the number of job level machine learning models and the risk output category of the job level machine learning models.

In step S6a, the determined risk is provided to the user as a result of the identification, the providing including displaying in a terminal interface, sending an email, sending a short message over a mobile communication network, or sending a social message to a user social media account.

Fig. 6 is a flowchart of a method of processing prediction sample data according to an embodiment of the present invention. In this embodiment, the method comprises the following steps:

step S1b, monitoring feedback information of the user for each recognition result. And monitoring feedback information of the user on the identification result after evaluating according to the recruitment interview information provided by the user to identify whether the recruitment faced by the current user is a false recruitment and providing the identification result to the user. In one embodiment, confirmation information of "at risk" or "no risk" in the recognition result, evaluation information of "correct"/"error" of the recognition result, and the like may be included in the feedback information.

In step S2b, the confirmation information of the user on the recognition result is extracted. From the information fed back in step S1b, confirmation information of "at risk" or "no risk", or evaluation information of "correct"/"incorrect" is extracted.

And step S3b, setting corresponding labels for the corresponding prediction samples according to the confirmation information of the user. For example, when the user confirms that the recruiting interview input by the user is real recruiting, a real recruiting label is set for each prediction sample evaluated for recruiting, and if the user confirms that the recruiting interview input by the user is false recruiting, a false recruiting label is set for each prediction sample evaluated for recruiting. In a preferred embodiment, more information fields are included in the user's feedback information, such as a reason that the user is required to fill out when the user confirms a false recruitment. The invention analyzes the reasons filled by the user, evaluates false root causes from two aspects of 'enterprise' and 'position', and sets labels for the current prediction samples at all levels.

And step S4b, storing the prediction samples with the set labels into a training set, namely, storing the prediction samples with the set labels into a data set for training a model by the system, thereby enriching the training data.

Step S5b, determining whether a model update condition is reached, where the update condition is, for example, a preset update period, such as updating the model once per week/month, or counting the number of newly added training samples in the model training data set, and determining whether the number of newly added training samples reaches a threshold value. If the update period is reached or the number of newly added training samples reaches the threshold, it is determined that the model update condition is met, step S6b is executed, otherwise, the step S1b is executed.

And step S6b, optimizing and updating the machine learning model currently used by adopting the training data set.

The invention can continuously accumulate the training data without manually setting labels of the training data, and the optimization of the model can be automatically carried out without manual intervention, thereby saving a large amount of manpower and time; and with the optimization of the model, the accuracy of model evaluation is gradually improved, and the false recruitment can be identified more accurately, so that the benefit of the user is better maintained.

In an embodiment, the present invention further provides a method for false recruitment early warning, referring to fig. 7, where fig. 7 is a flowchart of a method for false recruitment early warning according to an embodiment of the present invention, including:

and step S1c, determining whether or not the recruitment interview information provided by the user is received, executing step S2c if the recruitment interview information provided by the user is received, and repeating the step if the recruitment interview information is not received.

And step S2c, extracting the target recruitment enterprise information, the target position information and the interview information from the recruitment interview information provided by the user.

And step S3c, information crawling is carried out in the network. And respectively crawling information in the network according to a plurality of indexes in the enterprise level, the job level and the interview level based on the extracted target recruitment enterprise information, the target job information and the interview information.

In step S4c, the crawled information is processed into index data of a corresponding level.

Step S5c, risk assessment is performed based on the index data of each level. For example, risk assessment according to the procedure shown in any of FIGS. 2-5.

And S6c, backtracking and analyzing index data corresponding to the risks based on the evaluated risks to obtain early warning information and providing the early warning information for the user. Wherein the early warning information comprises problem index data causing risks and risk content.

The steps of backtracking and analyzing the index data corresponding to the risk include:

firstly, according to the monitored risk types, traversing the index data used in the risk assessment to calculate the contribution degree of each index data to the assessed risk.

Then, the indexes are sorted according to the contribution degree to the risk, and the index with the highest sorting or the index with the contribution degree larger than a threshold value is determined as the abnormal index.

And finally, generating risk content based on the content of the abnormal index and/or the incidence relation of the contents of the abnormal indexes.

Wherein, a feature importance determination method in feature engineering can be adopted to determine the contribution degree of the index data to the evaluated risk. The feature importance determination method is, for example, an expert conference method, in which experts in the field designate and identify the importance of each index used in the present invention, and for example, a rough set theory, an information entropy, and other methods are used to identify the importance of each index used in the present invention, or the importance of each index is determined by mining the association relationship between the index and the output through a data mining technology. Based on the importance of the indexes determined by the methods, when the indexes are traced back, the contribution degree is determined by inquiring the importance identification corresponding to each index.

There are other ways of determining the importance of features than those described above. For example, when a machine learning model of a neural network is used, since the weights of hidden nodes in the neural network represent the importance of the features of the corresponding input layer nodes, when the machine learning model is the neural network model, the weights of the hidden nodes are read, and the contribution degree of the index can be determined according to the weights of the hidden nodes.

For another example, for a machine learning model adopting a decision tree or a random forest algorithm, according to a decision tree generation principle, the sequence of the features selected in the process of dividing the decision tree can be used as an importance ranking of the features, and the ranking can be known through feature _ importances _ attribute in sklern, so that when the risk is backtracked, the importance of each index used for evaluating the risk, namely the contribution degree, can be obtained by calling the feature _ importances _ attribute.

After the contribution degrees of the indexes are determined, a plurality of indexes (such as the first three indexes) with the highest contribution degree sequence can be used as abnormal indexes, and after the abnormal indexes are determined, the contents of the abnormal indexes are read, so that points which possibly generate risks and specific risk contents are positioned. And analyzing whether the contents of the three abnormal indexes have a correlation relationship, and if so, taking the correlated contents as risk contents.

The invention is preset with risk events and corresponding solutions thereof which can be generated corresponding to various risk contents. Therefore, after the risk content is determined, the database is queried according to the risk content, and a coping plan matching the risk content is determined.

Taking the evaluation of the enterprise risk as an example, if the "enterprise lawsuit event" is found to be the main cause of the enterprise risk after backtracking, the corresponding scheme provided by the system includes: (1) the HR can be inquired and verified appropriately according to the enterprise labor dispute litigation situation in interviewing; (2) before interviewing or entering into work, please collect relevant information on the internet to understand the details of related litigation, and to enhance the overall understanding of the problem.

And adding the risk content and the corresponding scheme into the early warning information and presenting the early warning information to a user together. In addition, in order to enable the user to know the identified content more or facilitate the user to inquire in the future, the method and the system can also generate a risk report for the content in the early warning information and store or provide the risk report for the user. For example, the estimated risk level, risk content and corresponding plan generation risk report are sent to the user or stored in the user account.

Fig. 8 is a functional block diagram of a false recruitment identification system according to one embodiment of the present invention. In this embodiment, the system for identifying the false recruitment comprises a user information acquisition module 1, a data collection module 2, an index data generation module 3 and a risk assessment module 4. The user information acquisition module 1 is connected with the data collection module 2, and the user information acquisition module 1 receives recruitment interview information provided by a user and extracts target recruitment enterprise information, target position information and interview information from the user recruitment interview information. The data collection module 2 is connected to the internet, connected to the user information acquisition module 1, and configured to perform information crawling in the network according to multiple indexes in an enterprise level, a job level, and an interview level, based on the extracted target recruitment enterprise information, target job position information, and interview information, to obtain related information such as a recruitment enterprise, a target job position, and the like. The index data generation module 3 is connected with the data collection module 2, and processes the crawled information of the enterprise level, the job level and the interview level to obtain a plurality of index data of corresponding levels, wherein the index data is normalized into specific index data based on the obtained information of each level, such as floating point number, character string vector or certain code, through the processing of the index data generation module 3. The risk assessment module 4 is connected to the index data generation module 3, and configured to assess the target enterprise according to a plurality of index data of each level to obtain enterprise-level risk, job-level risk and interview-level risk of the target enterprise for the virtual recruitment.

Fig. 9 is a functional block diagram of a portion of a false recruitment identification system according to another embodiment of the present invention. The present embodiment employs a machine learning model to assess risk. Therefore, in this embodiment, in addition to the modules in fig. 8, the prediction sample generation module 5 is further included, and is connected to the index data generation module 3, and based on the samples required by the machine learning models of each stage, each index data is normalized into feature data, and the feature data corresponding to the index data of each stage are combined together to form the prediction sample of the machine learning model of each stage. . The risk assessment module 4 in the present embodiment includes an enterprise-level risk assessment unit 41, a job-level risk assessment unit 42, an interview-level risk assessment unit 43, an encoding unit 44, and a risk level query unit 45. The prediction sample generation module 5 inputs the prediction samples of each level to the enterprise level risk assessment unit 41, the job level risk assessment unit 42 and the interview level risk assessment unit 43, and the enterprise level risk assessment unit 41 obtains the enterprise level risk probability by inputting the prediction samples of each level to the trained enterprise level machine learning model. The job-level risk assessment unit 42 obtains the job-level risk probability by inputting job-level prediction samples to the trained job-level machine learning model. The interview-level risk assessment unit 43 obtains interview-level risk probability by inputting interview-level prediction samples to the trained interview-level machine learning model. In one embodiment, the three units may directly provide the obtained risk probability to the user as the identification result, in this embodiment, the three units output the obtained risk probability to the encoding unit 44, the encoding unit 44 encodes the obtained risk levels of the enterprise level, the job level and the interview level to obtain a risk code, and outputs the risk code to the risk level query unit 45, and the risk level query unit 45 queries a correspondence table between the risk code and the risk level, such as table 1 referred to in the foregoing description of the method, according to the risk code, and determines the risk level corresponding to the risk code as the final risk level. Of course, after obtaining the risk code, the encoding unit 44 may also calculate a weighted sum of the risk codes according to the weights of the risks at each level, and the risk level query unit 45 queries a table of correspondence between the weighted sum of the risk codes and the risk level, such as the table 2 mentioned in the above description of the method, and determines the risk level corresponding to the weighted sum of the risk codes as the final risk level.

Fig. 10 is a functional block diagram of a false recruitment identification system according to yet another embodiment of the present invention. In contrast to the embodiment shown in fig. 9, the risk assessment module 4 in this embodiment further includes a selection unit 46 in addition to the enterprise level risk assessment unit 41, the job level risk assessment unit 42, and the interview level risk assessment unit 43. In the present embodiment, the machine learning models used by the three evaluation units have an association with each other. According to the sequence of the enterprise level, the position level and the interview level from top to bottom, the next-level machine learning model is obtained by training according to training data corresponding to the risk level output by the previous-level machine learning model, therefore, the number of the next-level machine learning model is the same as that of the risk level of the previous-level machine learning model, the risk assessment is also performed step by step from top to bottom, and the next-level machine learning model is required to be selected according to the risk level of the previous-level machine learning model.

Specifically, the enterprise-level risk assessment unit 41 receives the enterprise-level prediction samples obtained by the prediction sample generation module 5, and notifies the selection unit 46, and the selection unit 46 selects an enterprise-level machine learning model from the model library and sends the model to the enterprise-level risk assessment unit 41. The enterprise-level risk assessment unit 41 inputs the enterprise-level prediction samples into the enterprise-level machine learning model, obtains enterprise-level risk levels through the enterprise-level machine learning model assessment, and notifies the job-level risk assessment unit 42 while sending the enterprise-level risk levels to the selection unit 46. The selection unit 46 selects an appropriate job-level machine learning model according to the enterprise-level risk level and sends the selected model to the job-level risk assessment unit 42. After receiving the notification from the enterprise-level risk assessment unit 41 and the job-level machine learning model sent by the selection unit 46, the job-level risk assessment unit 42 inputs the job-level prediction samples received from the prediction sample generation module 5 into the job-level machine learning model, assesses the job-level risk level by the job-level machine learning model to obtain the job-level risk level, outputs the job-level risk level to the selection unit 46, and outputs the notification to the interview-level risk assessment unit 43. The selecting unit 46 selects the corresponding interview-level machine learning model according to the position-level risk level and sends the interview-level risk model to the interview-level risk evaluating unit 43. After receiving the interview-level machine learning model sent by the notification and selection unit 46 of the job-level risk assessment unit 42, the interview-level risk assessment unit 43 inputs the interview-level prediction samples received from the prediction sample generation module 5 to the interview-level machine learning model, and obtains the final level risk level through assessment by the interview-level machine learning model.

The enterprise-level machine learning model and the job-level machine learning model outputs in the foregoing embodiments are classified into two categories, respectively defined as two risk levels, "at risk" and "no risk". The output of the interview-level machine learning model is classified into two categories or more categories, wherein the two categories are respectively defined as two risk levels of 'risk' and 'no risk', and the more categories are respectively defined as risk levels with different degrees from none to many.

FIG. 11 is a functional block diagram of a subscriber information acquisition module according to one embodiment of the present invention. In this embodiment, the user information acquisition module includes a messaging unit 11 and an information extraction unit 12. The messaging unit 11 is used as an interactive interface between the system and the user, and on one hand, is connected with the risk assessment module 4 to send the final recognition result to the user, and on the other hand, receives the recruitment interview information provided by the user and outputs the recruitment interview information to the information extraction unit 12. The information extraction unit 12 extracts target recruitment enterprise information, target position information, and interview information from the recruitment interview information provided by the user.

Wherein the messaging unit 11 comprises one or more of the following: an application terminal user interaction unit 110, a mail processing unit 111, a mobile short message processing unit 112 and a social media message processing unit 113. The application terminal user interaction unit 110 at least includes an input interface through which recruitment interview information provided by a user via the application terminal can be acquired, and in addition, the application terminal user interaction unit 110 may also include a display interface for displaying information, such as information of identified risk level, risk content or corresponding scheme. The mail processing unit 111 may identify the recruiting interview information provided by the user via a mail manner according to the mail address or the subject, or may send a message to the user via a mail manner. The mobile short message processing unit 112 identifies the recruitment interview information provided by the user in the short message mode according to the message sending number and the theme in the short message received by the mobile communication network, and can also send the message to the user. The social media message processing unit 113 identifies recruitment interview information provided by the user in a social media manner from the social media message according to the message sender or sends the message to the user.

FIG. 12 is a functional block diagram of a data processing system according to one embodiment of the present invention. The data processing system in this embodiment includes a user information obtaining module 1, a data collecting module 2, an index data generating module 3, and a prediction sample generating module 5, that is, some of the modules in the false recruitment identification system in fig. 9 or fig. 10 constitute a data processing system, and perform information extraction, information crawling, data rule range processing, and the like based on recruitment interview information provided by a user, so as to obtain a prediction sample for the machine learning model to evaluate risk. In particular, FIG. 13 is a functional block diagram of a data collection module according to one embodiment of the present invention. In the present embodiment, the data collection module includes an index acquisition unit 21, an index analysis unit 22, an information crawling unit 23, and an information matrix construction unit 24.

The index obtaining unit 21 is configured to read a plurality of indexes respectively applied to an enterprise level, a job level, and an interview level. In the embodiment, indexes for evaluating risks at different levels are stored in the system, and the index acquisition unit 21 reads the indexes from the system database and sends the indexes to the index analysis unit 22. The index analyzing unit 22 analyzes each index, determines index reference contents required for obtaining the index data, and sends the index reference contents to the information crawling unit 23. The information crawling unit 23 is connected with the index analysis unit 22, and crawls corresponding information from the internet according to the determined index reference content. In one embodiment, one or more retrieval keywords corresponding to the index are stored in the system, for example, when the index obtaining unit 21 reads the index of "number of enterprises", the index analyzing unit 22 queries the retrieval keywords to obtain the index reference content "number/quantity of enterprises", and the information crawling unit 23 searches and obtains the information of the number of enterprises in the target recruitment enterprise official website according to the index reference content "number/quantity of enterprises" to obtain the information meeting the index of "number of enterprises". For another example, when the index acquisition unit 21 reads the enterprise-level index "enterprise legal action event", the index analysis unit 22 queries the search keyword to obtain the index reference content "labor/defendant", and the information crawling unit 23 queries related action information of the target recruitment enterprise in the recent period (for example, five years) from the content.

In this embodiment, the information matrix constructing unit 24 is connected to the information crawling unit 23, and establishes an information matrix according to the association between the crawled information. The information matrix comprises one or more pieces of position information released by the target enterprise and recruitment platform qualification information for releasing the position information; the job position information released by one or more similar enterprises similar to the target enterprise and the recruitment platform qualification information for releasing the job position information. The positions published by the target enterprise comprise a target position and a second position different from the target position; the recruitment platforms include one or more target recruitment platforms that publish a target position provided by a target enterprise and one or more second recruitment platforms that publish second position information different from the target position.

FIG. 14 is a functional block diagram of an index data generation module according to an embodiment of the present invention. In the present embodiment, the index data generation module includes a data cleansing unit 31, a single index extraction unit 32, and a composite index calculation unit 33. The data cleaning unit 31 performs data cleaning on the crawled original information, including removing some network symbols and punctuation marks, querying a stop word list to remove stop words, and the like. The single index extraction unit 32 is connected to the data cleaning unit 31, and extracts and normalizes single index data from the cleaned data. The single index is, for example, some indexes that do not require complicated calculation and processing, such as indexes of "number of business persons", "number of branch companies", "registered fund", "job title", "academic requirement", and the like. For the indexes, index data can be extracted by identifying keywords and processed into character string vectors, floating point numerical values or codes according to specific indexes. The composite index calculation unit 33 is connected to the single index extraction unit 32, and is configured to calculate more than one single index data according to a composite index calculation rule to obtain composite index data. The composite index is, for example, an index such as "degree of introduction of recruitment platform", "qualification grade of recruitment platform", and "external consistency coefficient of target position". The calculation method is as described in the above description of the method, and is not described herein again.

FIG. 15 is a functional block diagram of a model training module in a data processing system according to another embodiment of the present invention. The model training module 6 in this embodiment comprises a training data set unit 61, a model training unit 62, a user feedback monitoring unit 63, a sample labeling unit 64 and a model updating unit 65. The training data set unit 61 is configured to provide a data set for training a model, and respectively includes an enterprise-level training data subset, a position-level training data subset, and an interview-level training data subset according to a type of the model to be trained. Further, each training data subset further comprises a training subset and a validation subset. The model training unit 62 performs model training with data in the training data subsets of the corresponding types according to the training model types to obtain an enterprise-level machine learning model, a position-level machine learning model, and an interview-level machine learning model, respectively. In one embodiment, the model training units 62 perform model training with independent training sets to obtain three independent machine learning models, and in another embodiment, the job-level training data subsets respectively include different subsets of training data with corresponding enterprise risks according to risk levels of the enterprise-level machine learning models, and different job-level machine learning models are obtained according to different training subsets. Similarly, the subset of the interview-level training data includes a plurality of training subsets formed by training data specifically corresponding to the enterprise risk level and the job level risk level, so that different interview-level machine learning models are obtained according to different training subsets. As mentioned above in connection with the example of the method according to the present invention, the enterprise-level machine learning model M1, the job-level machine learning models M2.1, M2.2, and the interview-level machine learning models M3.1, M3.2, M3.3, and M3.4 are obtained through different training subsets. In order to extend the training set, the user feedback monitoring unit 63 monitors feedback information of the user on the result of the false recruitment recognition, wherein the feedback information at least comprises risky or non-risky confirmation information. The sample labeling unit 64 is connected with the user feedback monitoring unit 63, and carries out risk labeling on the data of the obtained identification result based on the feedback information of the user, and adds the data to the corresponding training data subset, thereby achieving the purpose of enriching the training data. The model updating unit 65 is connected to the training data set unit 61, and is configured to monitor an updating condition, and send an updating notification to the model training unit 62 when the model updating condition is met. The model training unit 62 performs model training, optimization, and updating with training data. The update condition is, for example, that a preset update period is reached, such as updating the model once per week/month, or a newly added training sample reaches a threshold. Thus, the model updating unit 65 counts the time after each model update and transmits an update notification to the model training unit 62 when the count period has been reached. Or the model updating unit 65 counts the newly added training data in the training data set, and when the number of the newly added training data reaches a threshold value, informs the model training unit to optimize and update the original machine learning model with the current training data.

Fig. 16 is a functional block diagram of a false recruitment warning system according to one embodiment of the present invention. In this embodiment, the early warning system includes a user information obtaining module 1, a data collecting module 2, an index data generating module 3, a risk evaluating module 4, and an early warning module 7. The user information obtaining module 1, the data collecting module 2, the index data generating module 3, and the risk evaluating module 4 are the same as those in the virtual recruitment identification system in the foregoing embodiment, and are not described herein again.

The early warning module 7 is connected with the risk assessment module 4, and is used for backtracking and analyzing index data corresponding to the risk in response to the assessed risk to obtain early warning information and providing the early warning information for the user.

Fig. 17 is a functional block diagram of an early warning module according to an embodiment of the present invention. In this embodiment, the early warning module includes a risk monitoring unit 71, a risk content determining unit 72, an early warning information generating unit 73, and an early warning information sending unit 74.

The risk monitoring unit 71 is connected to the risk assessment module 4, monitors whether the risk assessment module 4 assesses a risk, and sends a notification to the risk content determination unit 72 when it is detected that the risk assessment module 4 assesses a risk. The risk content determination unit 72 traverses the index data for evaluating the risk based on the evaluation to the risk type to obtain abnormal index data, and determines the risk content based on the content of the one or more abnormal index data or the association relationship thereof. Any one of the foregoing methods may be adopted when determining the abnormal index data, and details are not repeated here. The warning information generating unit 73 is connected to the risk content determining unit 72, and generates warning information based on the risk type and the risk content. The warning information sending unit 74 is connected to the warning information generating unit 73, and is configured to provide the warning information to the user. The warning information sending unit 74 may adopt one or more of an interactive information sending unit, a mail processing unit, a mobile short message processing unit, and a social media message processing unit, that is, the warning information sending unit 74 may be combined with the user information obtaining module into one module, so that the module is implemented as a bidirectional function with both message receiving and sending functions, which is specifically referred to the foregoing embodiment and is not described herein again.

Fig. 18 is a functional block diagram of an early warning module according to another embodiment of the present invention. In this embodiment, compared to the embodiment of fig. 17, except that the prediction unit 75 is added, other modules have the same functions, and are not described herein again. The prediction unit 75 is connected to the risk content determination unit 72, and is configured to predict a risk event and a risk response scheme according to the risk content, and add the risk event and the risk response scheme to the warning information. For example, the prediction unit 75 queries various suggestions and corresponding schemes preset in the system database according to the risk content, thereby obtaining suggestions matched with the suggestions, and combines a plurality of suggestions and corresponding schemes together to add to the early warning information. For example, the system database may store various possible risk events corresponding to the risk content and corresponding solutions, and add these pieces of information to the warning information and transmit them to the user.

Fig. 19 is a functional block diagram of an early warning module according to another embodiment of the present invention. In this embodiment, compared with the embodiment of fig. 18, except that the risk report generating unit 76 is added, the functions of the other modules are the same, and are not described herein again. The risk report generating unit 76 is connected to the warning information generating unit 73, and is configured to generate a risk report according to the content in the warning information. Correspondingly, the early warning information sending unit stores the risk report in a preset position or sends the risk report to the user.

The above embodiments are provided only for illustrating the present invention and not for limiting the present invention, and those skilled in the art can make various changes and modifications without departing from the scope of the present invention, and therefore, all equivalent technical solutions should fall within the scope of the present invention.

Claims

1. A false recruitment early warning method comprising:

2. The method of claim 1, wherein the pre-warning information includes at least a risk type and/or a risk level, the risk type including enterprise level risk, job level risk, interview level risk, and/or final risk; the risk level includes one of a plurality of risk levels from small to large.

3. The method of claim 2, further comprising:

standardizing each index data into the characteristic data of the machine learning model, and combining the characteristic data corresponding to each level of index data together to form a prediction sample of each level of machine learning model; and

and inputting the prediction samples of all levels to the machine learning model of the corresponding level for risk assessment.

4. The method of claim 3, wherein the enterprise-level machine learning model is one and the job-level machine learning models are plural, each corresponding to a risk level output by the enterprise-level machine learning model; the interview level machine learning models are multiple and correspond to the risk levels output by the job level machine learning models respectively;

correspondingly, the step of performing a risk assessment further comprises: and selecting a next-level machine learning model according to the risk level output by the previous-level machine learning model from top to bottom according to the sequence of the enterprise level, the job level and the interview level, wherein the risk and the level output by the interview-level machine learning model are the final risk and level.

5. The method of claim 3, wherein the enterprise level machine learning model, the job level machine learning model, and the interview level machine learning model are each one and independent of each other;

correspondingly, the step of performing a risk assessment further comprises:

and respectively inputting the prediction samples of all levels into corresponding machine learning models, and respectively obtaining enterprise level risk types and risk levels, position level risk types and risk levels and interview level risk types and risk levels through evaluation of all the machine learning models.

6. The method of claim 5, wherein the step of conducting a risk assessment further comprises:

coding the obtained risk levels of the enterprise level, the job level and the interview level to obtain a risk code;

inquiring a corresponding table of the risk codes and the risk grades according to the risk codes; and

determining a risk level corresponding to the risk code as a final risk level.

7. The method of claim 1, the step of backtracking and analyzing the indicator data corresponding to the risk comprising:

according to the monitored risk types, traversing a plurality of index data for evaluating the risks to obtain the contribution degree of each index data to the evaluated risks;

the indexes are sorted according to the contribution degree to the risk, and the indexes with the highest sorting or the indexes with the contribution degree larger than a threshold value are determined as abnormal indexes; and

and generating risk content based on the content of the abnormal index and/or the incidence relation of the contents of the plurality of abnormal indexes.

8. The method of claim 7, further comprising:

and determining a risk coping scheme based on the risk content, and adding the risk content and the coping scheme thereof to early warning information.

9. The method of claim 1 or 2 or 7 or 8, further comprising: and generating a risk report according to the content in the early warning information, storing the risk report in a preset position or providing the risk report to the user.

10. The method of claim 1, wherein pre-alert information is provided to the user by one or more of: displaying in a terminal interface, sending an email, sending a short message through a mobile communication network, and sending a social message through a user social media account.

11. A false recruitment warning system comprising:

the system comprises a user information acquisition module, a data processing module and a data processing module, wherein the user information acquisition module is configured to receive recruitment interview information provided by a user and extract target recruitment enterprise information, target position information and interview information from the user recruitment interview information;

a data collection module connected to the internet, connected to the user information acquisition module, and configured to perform information crawling in the network according to a plurality of indexes in an enterprise level, a job level, and an interview level, respectively, based on the extracted target recruitment enterprise information, target job position information, and interview information;

the index data generation module is connected with the data collection module and is used for processing the crawled information of the enterprise level, the job level and the interview level to obtain a plurality of index data of corresponding levels; and

a risk assessment module connected with the index data generation module and configured to perform risk assessment from an enterprise level, a job level and an interview level according to the index data of each level according to a risk assessment policy; and

and the early warning module is connected with the risk assessment module and is configured to respond to the assessed risk, backtrack and analyze index data corresponding to the risk to obtain early warning information and provide the early warning information for the user.

12. The system of claim 11, wherein the early warning module comprises:

a risk monitoring unit connected with the risk assessment module and configured to monitor whether the risk assessment module assesses a risk; and

a risk content determination unit, connected to the risk monitoring unit, configured to backtrack and analyze the index data corresponding to the risk based on the assessed risk type to obtain one or more abnormal indexes, and determine a risk content based on the content of the one or more abnormal index data and the association thereof;

an early warning information generation unit connected with the risk content determination unit and configured to generate early warning information based on a risk type and a risk content; and

an early warning information sending unit connected with the early warning information generating unit and configured to provide the early warning information to the user.

13. The system of claim 12, wherein the early warning module further comprises a prediction unit coupled to the risk content determination unit and configured to determine a risk response scenario from risk content and add the risk response scenario to early warning information.

14. The system of claim 13, wherein the early warning module further comprises a risk report generating unit, connected to the early warning information generating unit, configured to generate a risk report according to content in early warning information; correspondingly, the early warning information sending unit stores the risk report in a preset position or sends the risk report to the user.

15. The system of claim 14, wherein the warning information sending unit is one or more of the following:

the interactive information sending unit is configured to process the early warning information into display information according to the format requirement of the user interactive interface and send the display information to the user interactive interface for display;

the mail processing unit is configured to send early warning information to the user in a mail mode based on a preset mail address and/or a preset subject;

a mobile short message processing unit configured to send the warning information to the user in a short message manner via a mobile communication network;

the social media message processing unit is configured to send early warning information to the user in a social media mode.