CN106022708A - Method for predicting employee resignation - Google Patents
Method for predicting employee resignation Download PDFInfo
- Publication number
- CN106022708A CN106022708A CN201610300752.9A CN201610300752A CN106022708A CN 106022708 A CN106022708 A CN 106022708A CN 201610300752 A CN201610300752 A CN 201610300752A CN 106022708 A CN106022708 A CN 106022708A
- Authority
- CN
- China
- Prior art keywords
- user behavior
- behavior data
- characteristic vector
- leaving office
- predicted
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/10—Office automation; Time management
- G06Q10/105—Human resources
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Strategic Management (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Tourism & Hospitality (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method for predicting employee resignation. User behavior data, corresponding to a preset resignation attribute entry, of a training sample is acquired, based on the obtained user behavior data, feature vectors of the training sample are extracted, and based on the extracted feature vectors, a resignation prediction model used for predicting whether employees to be predicated have resignation intensions is trained. The technical problem of how to predict the employee resignation is solved, whether the employees to be predicted have the resignation intentions can be predicted simply according to the user behavior data of the employees to be predicted, an enterprise can know whether the employees to be predicted have the resignation intentions at an early stage and takes corresponding measures to reduce the enterprise resignation rate, accordingly, the manpower or monetary cost spent by the enterprise on re-hiring is greatly saved, and normal operation or work progress of the enterprise is guaranteed.
Description
Technical field
The present invention relates to communication technical field, be specifically related to a kind of method predicting labor turnover.
Background technology
Although labor turnover phenomenon is customary in enterprise, but more or less enterprise can be owing to not knowing that employee has leaving office meaning in advance
To and be in the most passive circumstances.On the one hand, the technology outstanding for some or management personnel, enterprise can not close early
That manages pacifies or keeps;On the other hand, in the face of the unexpected leaving office of employee, enterprise may cannot immediately recruit suitable employee or
The personnel arranging corresponding post are operated handing-over.So needing a kind of method that can predict labor turnover of offer badly.
Summary of the invention
The invention provides a kind of method predicting labor turnover, to solve how to predict the technical problem of labor turnover.
According to an aspect of the present invention, it is provided that a kind of method predicting labor turnover, including:
Preset leaving office attributes entries;
Gathering the user behavior data corresponding with leaving office attributes entries of training sample, wherein, training sample includes turnover intention
Employee and the training sample without turnover intention employee;
Based on user behavior data, extract the characteristic vector of training sample;
Grader is trained, it is thus achieved that leaving office forecast model according to characteristic vector;
According to leaving office forecast model, determine whether employee to be predicted has turnover intention.
Further, leaving office attributes entries includes:
History chat data, job performance, job tenure, income level, the last time promote time interval, working distance, step on
One or more combinations in record recruitment and job hunting net frequency entries.
Further, the user behavior data corresponding with history chat data entry gathering training sample includes:
Gather SMS historical record and/or the instant messaging historical record of training sample, chatting with history as training sample
The user behavior data that Data Entry is corresponding.
Further, based on user behavior data, the characteristic vector extracting training sample includes:
Word frequency is used to obtain the characteristic vector of the user behavior data corresponding with history chat data entry against text algorithm;
According to predefined mark rule, the user row corresponding to other leaving office attributes entries in addition to history chat data entry
Quantitative identifying is carried out, it is thus achieved that the characteristic vector of the user behavior data that other leaving office attributes entries are corresponding for data;
Characteristic vector and other leaving office attributes entries according to the user behavior data corresponding with history chat data entry are corresponding
The characteristic vector of user behavior data, it is thus achieved that the characteristic vector of training sample.
Further, use word frequency against text algorithm obtain the feature of the user behavior data corresponding with history chat data entry to
Amount includes:
The user behavior data corresponding with history chat data entry is converted into the character string of text formatting, it is thus achieved that history chat literary composition
This;
History chat text is carried out participle, semantic disambiguation, removes stop words operation, it is thus achieved that participle text;
Word frequency is used to obtain the weighted value of the participle text mated in participle text with the leaving office Feature Words preset against text algorithm, and
Using weighted value as the characteristic vector of the user behavior data corresponding with history chat data entry.
Further, according to leaving office forecast model, determine whether employee to be predicted has turnover intention to include:
Gather the to be predicted user behavior data corresponding with leaving office attributes entries of employee to be predicted;
Based on user behavior data to be predicted, extract the characteristic vector of user behavior data to be predicted;
Characteristic vector according to user behavior data to be predicted and leaving office forecast model, determine whether employee to be predicted has leaving office meaning
To.
Further, grader includes:
Any one in support vector machine classifier, Bayes classifier, maximum entropy classifiers.
The method have the advantages that
The invention provides a kind of method predicting labor turnover, by gather training sample with leaving office attribute bar set in advance
The user behavior data that mesh is corresponding, and the characteristic vector of training sample is extracted based on the user behavior data obtained, and based on carrying
The eigen vector that takes is trained for predicting whether employee to be predicted has the leaving office forecast model of turnover intention, solves the how person of prediction
The technical problem that work is left office, it is achieved that just turnover intention whether can be had to carry out it according to the user behavior data of employee to be predicted pre-
Surveying, whether beneficially enterprise knows employee early a turnover intention, and takes corresponding measure to reduce enterprise's separation rate, thus significantly
Save enterprise and again recruit spent manpower or monetary cost and the normal operation having ensured enterprise or work progress.
In addition to objects, features and advantages described above, the present invention also has other objects, features and advantages.Below
Will be with reference to figure, the present invention is further detailed explanation.
Accompanying drawing explanation
The accompanying drawing of the part constituting the application is used for providing a further understanding of the present invention, the illustrative examples of the present invention and
Its explanation is used for explaining the present invention, is not intended that inappropriate limitation of the present invention.In the accompanying drawings:
Fig. 1 is the method flow diagram of the prediction labor turnover of the preferred embodiment of the present invention;
Fig. 2 be the preferred embodiment of the present invention for first simplify embodiment prediction labor turnover method flow diagram;
Fig. 3 be the preferred embodiment of the present invention for second simplify embodiment prediction labor turnover method flow diagram;
Fig. 4 be the preferred embodiment of the present invention for the 3rd simplify embodiment prediction labor turnover method flow diagram.
Detailed description of the invention
Below in conjunction with accompanying drawing, embodiments of the invention are described in detail, but the present invention can be defined by the claims and cover
Multitude of different ways implement.
With reference to Fig. 1, the preferred embodiments of the present invention provide a kind of method predicting labor turnover, including:
Step S101, presets leaving office attributes entries;
Step S102, gathers the user behavior data corresponding with leaving office attributes entries of training sample, and wherein, training sample includes
There are turnover intention employee and the training sample without turnover intention employee;
Step S103, based on user behavior data, extracts the characteristic vector of training sample;
Step S104, trains grader according to characteristic vector, it is thus achieved that leaving office forecast model;
Step S105, according to leaving office forecast model, determines whether employee to be predicted has turnover intention.
The invention provides a kind of method predicting labor turnover, by gather training sample with leaving office attribute bar set in advance
The user behavior data that mesh is corresponding, and the characteristic vector of training sample is extracted based on the user behavior data obtained, and based on carrying
The eigen vector that takes is trained for predicting whether employee to be predicted has the leaving office forecast model of turnover intention, solves the how person of prediction
The technical problem that work is left office, it is achieved that just turnover intention whether can be had to carry out it according to the user behavior data of employee to be predicted pre-
Surveying, whether beneficially enterprise knows employee early a turnover intention, and takes corresponding measure to reduce enterprise's separation rate, thus significantly
Save enterprise and again recruit spent manpower or monetary cost and the normal operation having ensured enterprise or work progress.
Whether available man-power resources employee has turnover intention, it is common that by carrying out, with employee, the result interviewed and combining employee
Usual work performance, carries out subjective forecast.The accuracy using this subjective forecast employee whether to have turnover intention is the highest, and
And the method for subjective forecast the most well promotes the suitability, namely whether prediction employee has turnover intention unified and objectively
Method, thus cause being required for individually being carried out subjective forecast by human resources for each employee, workload is relatively big, and efficiency is relatively
Low.
For this problem, whether prediction employee is had the problem of turnover intention to be converted to the classification problem in pattern recognition by the present embodiment.
Specifically, first the present embodiment trains for predicting whether employee to be predicted has the leaving office forecast model of turnover intention, leaves office pre-
The output result surveying model is divided into two kinds, is to have turnover intention and do not have turnover intention respectively, then pre-according to the leaving office trained
Survey whether employee to be predicted is had turnover intention to be predicted by model.In concrete implementation process, the present embodiment can have been chosen
The training sample of the employee through leaving office is as the training sample of the employee having turnover intention, and chooses the training sample of on-job employee
Training sample as the employee not having turnover intention.It should be noted that in order to ensure to train the leaving office forecast model tool obtained
Having of a relatively high precision of prediction, the quantity of the training sample that the present embodiment obtains should be big as far as possible, and for there being turnover intention
Should be suitable with the quantity of the training sample without turnover intention employee.
The present embodiment propose relatively newly user behavior data according to employee set up for predict employee whether have turnover intention from
Duty forecast model, and whether employee to be predicted has turnover intention to use this leaving office forecast model to predict, the most existing employing is subjective pre-
Survey whether employee has the accuracy of method of turnover intention higher, and whether have leaving office meaning by leaving office forecast model prediction employee
To predictive efficiency high, there is the bigger popularization suitability.
Alternatively, leaving office attributes entries includes: history chat data, job performance, job tenure, income level, nearest one
One or more combinations in secondary time interval of promoting, working distance, login recruitment and job hunting net frequency entries.
The existing many factors affecting labor turnover, such as job performance, job tenure, income level, the last time are promoted the time
Interval, working distance (the most also including distance, traffic time cost, change trains or buses number of times cost, expense cost etc.) etc. factor,
Therefore the present embodiment goes out from demission or other user behavior datas (such as history chat data or login recruitment and job hunting network data)
Send out, gather the user behavior data corresponding for each leaving office attributes entries respectively, and enter according to the user behavior data gathered
Row subsequent analysis.Certainly, the leaving office attributes entries in the present embodiment be not limited to above-mentioned these, such as can also include enterprise development
Entry, industry development entry etc..
The present embodiment according to existing life affects the factor of labor turnover or other user behavior datas (such as history chat data or
Log in recruitment and job hunting network data), leaving office attributes entries is set, it is achieved thereby that from each dimension user behavior number to training sample
According to being acquired, the accuracy and precision of prediction for improving forecast model provides important Data Source basis.
Alternatively, the user behavior data corresponding with history chat data entry gathering training sample includes:
Gather the SMS historical record of training sample and/or instant messaging historical record, chatting number with history as training sample
According to the user behavior data that entry is corresponding.
Specifically, in the present embodiment using the SMS historical record of training sample and/or instant messaging historical record as training sample
This user behavior data corresponding with history chat data entry.In actual implementation process, the present embodiment is not limited to only to
SMS historical record and/or instant messaging historical record are as the user row corresponding with history chat data entry of training sample
For data, such as can also obtain history chat data corresponding to the platform such as microblogging, forum as training sample and historical data
The user behavior data that entry is corresponding.
Alternatively, based on user behavior data, the characteristic vector extracting training sample includes:
Word frequency is used to obtain the characteristic vector of the user behavior data corresponding with history chat data entry against text algorithm;
According to predefined mark rule, the user row corresponding to other leaving office attributes entries in addition to history chat data entry
Quantitative identifying is carried out, it is thus achieved that the characteristic vector of the user behavior data that other leaving office attributes entries are corresponding for data;
Characteristic vector and other leaving office attributes entries according to the user behavior data corresponding with history chat data entry are corresponding
The characteristic vector of user behavior data, it is thus achieved that the characteristic vector of training sample.
The form of the user behavior data corresponding with leaving office attributes entries owing to gathering is different, particularly with history chat data
The user behavior data form difference that user behavior data form corresponding to entry is corresponding with other divorce attributes entries is bigger.Therefore this
Embodiment for the user behavior data corresponding with history chat data entry, and in addition to history chat data entry other from
User behavior data corresponding to duty attributes entries takes different characteristic vector pickup methods.
Specifically, when extracting the characteristic vector of the user behavior data corresponding with history chat data entry, the present embodiment takes word
The inverse text algorithm of frequency realizes.And extract the user behavior number corresponding with other leaving office attributes entries in addition to history chat data entry
According to characteristic vector time, first the present embodiment arranges mark rule, then leaves office to other in addition to history chat data entry
User behavior data corresponding to attributes entries carries out quantitative identifying, thus finally obtains user's row that other leaving office attributes entries are corresponding
Characteristic vector for data.Mark rule in the present embodiment by User Defined, such as when collect training sample with work
User behavior data corresponding to performance entry is " medium ", and the user behavior data corresponding with job tenure entry is " 3 years ",
Then can be respectively provided with corresponding quantitative identifying value according to the grade of job performance, such as, be " outstanding " etc. by job performance
It is " 80-100 " that level arranges corresponding quantitative identifying value scope, and " well " grade is corresponding " 60-79 ", " medium " grade
Corresponding " 40-59 ", the like, it is thus possible to the user behavior data corresponding with job performance entry obtained is carried out quantitative identifying.
Similarly, be " 3 years " when collecting the user behavior data corresponding with the job tenure entry of training sample, then can be according to work
The time making the term of office is respectively provided with corresponding quantitative identifying value, such as, arrange corresponding by job tenure for " 0-5 " year
Quantitative identifying value scope is " 80-100 ", is to arrange corresponding quantitative identifying value scope " 6-10 " year to be by job tenure
" 60-79 ", the rest may be inferred, it is thus possible to the user behavior data corresponding with job tenure is carried out quantitative identifying.It should be noted that
Quantitative identifying value that the present embodiment is arranged for user behavior data on-fixed, specifically the most self-defined by user.
The present embodiment, after the characteristic vector extracting user behavior data corresponding to different leaving office attributes entries, is combined
The characteristic vector of the training sample that rear acquisition is final.Specifically, when the user behavior that the different leaving office attributes entries obtained is corresponding
When the dimension of the characteristic vector of data is different, the dimension of the characteristic vector of different dimensions is converted into maximum by the present embodiment unification
The dimension of the characteristic vector of dimension is identical.Such as, when according to the user behavior data extraction corresponding with history chat data entry
The dimension of characteristic vector is 10, and the dimension of the characteristic vector extracted according to the user behavior data that other leaving office attributes entries are corresponding
When being respectively less than 10, then the dimension of the characteristic vector extracted by the user behavior data corresponding according to other leaving office attributes entries is all changed
Becoming 10 dimensions, the mode that specifically " 0 " can be used to fill completes.
In concrete implementation process, the present embodiment can also take other characteristic vector pickup modes to obtain the feature of training sample
Property vector, or identical characteristic vector pickup method can also be taked to extract the user that different leaving office attributes entries is corresponding simultaneously
The characteristic vector of behavioral data, specifically by User Defined.
The present embodiment is by taking different characteristic vector pickup sides to the user behavior data corresponding from different leaving office attributes entries
Formula, can take different characteristic vector pickup respectively in conjunction with the concrete form of user behavior data corresponding to different leaving office attributes entries
Mode, thus obtain the characteristic vector of the user behavior data corresponding with leaving office attributes entries, make characteristic vector and the leaving office of acquisition
User behavior data corresponding to attributes entries matches, and has more representativeness.Therefore the user that the present embodiment can not only will gather
Behavioral data carries out quantitative identifying, thus obtains the characteristic vector for train classification models of standard criterion, and by will be with
User behavior data corresponding to other leaving office attributes entries in addition to history chat data entry carries out quantitative identifying, makes full use of each
The user behavior data that type and multiple dimension obtain, thus establish unified and standard data basis for follow-up training grader.
Alternatively, word frequency is used to obtain the characteristic vector of the user behavior data corresponding with history chat data entry against text algorithm
Including:
The user behavior data corresponding with history chat data entry is converted into the character string of text formatting, it is thus achieved that history chat literary composition
This;
History chat text is carried out participle, semantic disambiguation, removes stop words operation, it is thus achieved that participle text;
Word frequency is used to obtain the weighted value of the participle text mated in participle text with the leaving office Feature Words preset against text algorithm, and
Using weighted value as the characteristic vector of the user behavior data corresponding with history chat data entry.
Specifically, the present embodiment is set in advance in chat process the leaving office Feature Words list being embodied with turnover intention, such as, " change
Work ", " job hunting ", " recruitment ", " looking for a job ", " leaving office ", " resignation " etc., use word frequency to obtain against text algorithm the most again
The weighted value of participle text that mates with the leaving office Feature Words preset in participle text, and using weighted value as with history chat number
Characteristic vector according to user behavior data corresponding to entry.Wherein, the word frequency of the present embodiment-inverse text algorithm uses TF-IDF function
The computing formula of the weighted value calculating the participle text mated in participle text with the leaving office Feature Words preset is:
w(tk,Tj)=tf (tk,Tj)×idf(tk),
Wherein w (tk,Tj) it is history chat text TjIn with preset leaving office Feature Words tkThe weighted value of the participle text of coupling,
tf(tk,Tj) it is tkIn history chat text TjIn word frequency number;Represent tkInverse in training set
Text frequency, N is history chat text total number, N in training sampleKT is comprised for the history chat text in training samplekGo through
Number.
Alternatively, according to leaving office forecast model, determine whether employee to be predicted has turnover intention to include:
Gather the to be predicted user behavior data corresponding with leaving office attributes entries of employee to be predicted;
Based on user behavior data to be predicted, extract the characteristic vector of user behavior data to be predicted;
Characteristic vector according to user behavior data to be predicted and leaving office forecast model, determine whether employee to be predicted has leaving office meaning
To.
Specifically, when needs predict whether employee to be predicted has turnover intention, the present embodiment first gather employee to be predicted with
The user behavior data to be predicted that leaving office attributes entries is corresponding, is then based on user behavior data to be predicted, extracts user to be predicted
The characteristic vector of behavioral data, and extract user behavior data to be predicted characteristic vector mode and training grader before extract instruction
The method of the characteristic vector practicing sample is consistent, prediction of finally the characteristic vector input of the user behavior data to be predicted extracted being left office
Model, and judge whether employee to be predicted has turnover intention according to the output result of the leaving office forecast model trained.
Alternatively, grader includes: any one in support vector machine classifier, Bayes classifier, maximum entropy classifiers.
It should be noted that the disaggregated model of the present embodiment training in advance is not limited to include svm classifier model, Bayes's classification mould
Type, maximum entropy disaggregated model, namely the present embodiment can also use the disaggregated model that other training in advance are good as prediction employee to be
The no forecast model having turnover intention.
With three embodiments simplified, the method for the present embodiment prediction labor turnover is carried out further specific description below.
Simplify embodiment one
With reference to Fig. 2, the method for the present embodiment prediction labor turnover includes:
Step S201, presets leaving office attributes entries.
Specifically, present embodiment assumes that the leaving office attributes entries only one of which of setting, specially history chat data entry.
Step S202, gathers the user behavior data corresponding with leaving office attributes entries of training sample, and wherein, training sample includes
There are turnover intention employee and the training sample without turnover intention employee.
Specifically, the present embodiment gathers chatting with history of the training sample of ex-employee and the training sample of in-service employee respectively
The user behavior data that Data Entry is corresponding.Namely gather SMS historical record and/or the instant messaging history note of training sample
Record, as the user behavior data corresponding with history chat data entry of training sample.In concrete implementation process, this reality
Execute example to obtain and treat training sample SMS historical record in certain time period and/or instant messaging historical record, such as
SMS in SMS historical record in one month recently and/or instant messaging historical record, or half a year recently is gone through
Records of the Historian record and/or instant messaging historical record etc., have by User Defined.
Step S203, use word frequency against text algorithm obtain the feature of the user behavior data corresponding with history chat data entry to
Amount.
Specifically, the present embodiment uses word frequency to obtain the user behavior data corresponding with history chat data entry against text algorithm
Characteristic vector includes:
Step S2031, is converted into the character string of text formatting by the user behavior data corresponding with history chat data entry, it is thus achieved that
History chat text.Specifically, the user behavior data corresponding with history chat data entry gathered due to the present embodiment may
Including various ways, such as text, picture, video, audio frequency, voice etc., thus corresponding with history chat number getting
User behavior data after, first convert thereof into the character string of text formatting, thus be subsequent extracted and history chat data entry
The characteristic vector of corresponding user behavior data lays the foundation.
Step S2032, carries out participle, semantic disambiguation, removes stop words operation, it is thus achieved that participle text history chat text.?
In concrete implementation process, the present embodiment carries out pretreatment to history chat text, thus obtains participle text, is not limited to only wrap
Include participle, semantic disambiguation, remove stop words operation, such as, can also include the operations such as part-of-speech tagging.And the present embodiment is to history
Chat text carries out the method for participle can use the multiple segmenting method such as maximum forward matching method or maximum reverse matching method.
Step S2033, uses word frequency to obtain the participle text that mates in participle text with default leaving office Feature Words against text algorithm
Weighted value, and using weighted value as the characteristic vector of the user behavior data corresponding with history chat data entry.
Specifically, present embodiment assumes that the leaving office Feature Words list that pre-sets for " changing jobs ", " job hunting ", " recruitment ", " look for
Work ", " leaving office ", " resignation ", use word frequency to obtain in participle text against text algorithm and default leaving office Feature Words the most again
The weighted value of participle text of coupling, and using weighted value as the feature of the user behavior data corresponding with history chat data entry
Vector.Namely the present embodiment add up respectively the participle text corresponding with history chat text comprises in leaving office Feature Words list from
The weighted value of duty Feature Words, for example, it is assumed that the present embodiment statistics and history chat text TjCorresponding participle text comprises leaving office
The computing formula of the weighted value of Feature Words (" changing jobs ") is:
w(tk,Tj)=tf (tk,Tj)×idf(tk),
Wherein w (tk,Tj) it is history chat text TjIn with preset leaving office Feature Words (" changing jobs ") tkThe participle text of coupling
Weighted value, tf (tk,Tj) it is tkIn history chat text TjIn word frequency number, namely history chat text TjMiddle appearance leaving office feature
The word frequency number that word " changes jobs ";Represent tkInverse text frequency in training set, N is instruction
Practice history chat text total number in sample, NKLeaving office Feature Words (" changing jobs ") is comprised for the history chat text in training sample
tkCount mesh one by one.According to above-mentioned formula, it is not difficult to calculate in history chat text and each spy that leaves office in leaving office Feature Words list
Levy the weighted value that word is the most corresponding, it is assumed that the present embodiment gets history chat text TjIn with leaving office Feature Words list for { " to change work
Make ", " job hunting ", " recruitment ", " looking for a job ", " leaving office ", " resignation " in leaving office Feature Words respectively correspondence weighted value be
w(t1,Tj)~w (t6,Tj), then the present embodiment is by { w (t1,Tj)、w(t2,Tj)、w(t3,Tj)、w(t4,Tj)、w(t5,Tj)、
w(t6,Tj) as training sample TjThe characteristic vector of the user behavior data corresponding with history chat data entry.
Step S204, trains grader according to characteristic vector, it is thus achieved that leaving office forecast model.Specifically, it is assumed that the instruction of the present embodiment
Practicing total sample number is N, the most respectively the characteristic vector of each training sample is inputted grader and is trained, thus obtains leaving office
Forecast model, it should be noted that in order to obtain of a relatively high classification accuracy and precision of prediction, the instruction that the present embodiment is chosen
The quantity practicing sample should be tried one's best greatly.
Step S205, according to leaving office forecast model, determines whether employee to be predicted has turnover intention.
Specifically, when needs predict whether employee to be predicted has turnover intention, the present embodiment first gather employee to be predicted with
The user behavior data to be predicted that leaving office attributes entries is corresponding, i.e. gathers the corresponding with history chat data entry of employee to be predicted
User behavior data, namely the SMS historical record of employee to be predicted and/or instant messaging historical record;It is then based on treating pre-
Survey user behavior data, extract the characteristic vector of user behavior data to be predicted, and extract the feature of user behavior data to be predicted
The mode of vector is consistent with the method for the characteristic vector extracting training sample before training grader, the user to be predicted that finally will extract
The characteristic vector input leaving office forecast model of behavioral data, and treat pre-according to the output result judgement of the leaving office forecast model trained
Survey whether employee has turnover intention.
The present embodiment is by obtaining the user behavior data corresponding with history chat data entry of employee to be predicted namely to be predicted
The SMS historical record of employee and/or instant messaging historical record, and extracted and history chat against text algorithm by word frequency
The characteristic vector of the user behavior data that Data Entry is corresponding, trains for predicting whether employee has the leaving office of turnover intention to predict
Model, solves the technical problem how predicting labor turnover, it is achieved that according to the SMS historical record of employee to be predicted and/
Or whether it just can be had turnover intention to be predicted by instant messaging historical record, beneficially enterprise know early employee whether have from
Duty purpose, and take corresponding measure to reduce enterprise's separation rate, thus be greatly saved enterprise and again recruit spent manpower or gold
Money cost and ensured normal operation or the work progress of enterprise.
Simplify embodiment two
With reference to Fig. 3, the method for the present embodiment prediction labor turnover includes:
Step S301, presets leaving office attributes entries.
Specifically, present embodiment assumes that the leaving office attributes entries of setting includes 5, respectively job performance entry, job tenure
Entry, income level entry, working distance entry, login recruitment and job hunting net frequency entries.
Step S302, gathers the user behavior data corresponding with leaving office attributes entries of training sample, and wherein, training sample includes
There are turnover intention employee and the training sample without turnover intention employee.
Specifically, the present embodiment gather respectively the training sample of ex-employee and the training sample of in-service employee with above-mentioned five
The user behavior data that leaving office attributes entries is corresponding.Assume that the present embodiment gathers corresponding for above-mentioned five leaving office attributes entries
User behavior data is as shown in table 1:
Table 1
Step S303, according to predefined mark rule, to other leaving office attributes entries pair in addition to history chat data entry
The user behavior data answered carries out quantitative identifying, it is thus achieved that the characteristic vector of the user behavior data that other leaving office attributes entries are corresponding.
Specifically, owing to the leaving office attributes entries in the present embodiment does not include history chat data entry, therefore according to predefined
Mark rule, carries out quantitative identifying to the user behavior data that other leaving office attributes entries in addition to history chat data entry are corresponding.
The present embodiment carries out quantitative identifying for the user behavior data that other leaving office attributes entries in addition to history chat data entry are corresponding
Mark rule by User Defined, the present embodiment, in order to unify quantitative identifying scope, will be carried out quantitatively for user behavior data
The scope of mark is arranged between scope 0-100, referring in particular to table 2.
Table 2
According to table 2, present embodiment assumes that the user behavior data obtained for table 1 obtains after carrying out quantitative identifying with five leaving office
The ident value of the user behavior data that attributes entries is corresponding be respectively 50,95,65,59,70}, due to the leaving office attributes entries of the present embodiment
Do not include history chat data entry, then directly by vector, { 50,95,65,59,70} as the characteristic vector of training sample.
Step S304, trains grader according to characteristic vector, it is thus achieved that leaving office forecast model.Specifically, it is assumed that the instruction of the present embodiment
Practicing total sample number is N, the most respectively the characteristic vector of each training sample is inputted grader and is trained, thus obtains leaving office prediction
Model, it should be noted that in order to obtain of a relatively high classification accuracy and precision of prediction, the training sample that the present embodiment is chosen
Quantity should be tried one's best greatly.
Step S305, according to leaving office forecast model, determines whether employee to be predicted has turnover intention.
Specifically, when needs predict whether employee to be predicted has turnover intention, the present embodiment first gather employee to be predicted with
The user behavior data to be predicted that leaving office attributes entries is corresponding, i.e. gather employee to be predicted with step S301 in set five from
The user behavior data that duty attributes entries is corresponding;It is then based on user behavior data to be predicted, extracts user behavior data to be predicted
Characteristic vector, and extract training sample before extracting the mode of the characteristic vector of user behavior data to be predicted and training grader
The method of characteristic vector is consistent, finally the characteristic vector of the user behavior data to be predicted extracted is inputted leaving office forecast model, and
Output result according to the leaving office forecast model trained judges whether employee to be predicted has turnover intention.
The present embodiment is by obtaining the user behavior data corresponding with leaving office attributes entries of employee to be predicted, and passes through to determine in advance
The user behavior data that other leaving office attributes entries in addition to history chat data entry are corresponding is quantitatively marked by the mark rule of justice
Know, thus obtain the characteristic vector of user behavior data corresponding to other leaving office attributes entries, and based on the user behavior number obtained
According to characteristic vector train for predicting whether employee has the leaving office forecast model of turnover intention, solve how to predict employee from
The technical problem of duty, it is achieved that just turnover intention whether can be had to be predicted it according to the user behavior data of employee to be predicted,
Be conducive to enterprise to know whether employee has turnover intention early, and take corresponding measure to reduce enterprise's separation rate, thus be greatly saved
Enterprise recruits spent manpower or monetary cost and the normal operation having ensured enterprise or work progress again.Additionally, this
The user behavior data of collection is carried out quantitative identifying by embodiment, it is hereby achieved that standard criterion for train classification models
Characteristic vector, and by arranging multiple leaving office attributes entries, from multiple dimensions, user behavior data can be carried out data acquisition,
It is favorably improved accuracy and the precision of prediction of leaving office forecast model of disaggregated model.
Simplify embodiment three
With reference to Fig. 4, the method for the present embodiment prediction labor turnover includes:
Step S401, presets leaving office attributes entries.
Specifically, present embodiment assumes that the leaving office attributes entries of setting includes 6, respectively history chat data entry, work
Performance entry, job tenure entry, income level entry, working distance entry, login recruitment and job hunting net frequency entries.
Step S402, gathers the user behavior data corresponding with leaving office attributes entries of training sample, and wherein, training sample includes
There are turnover intention employee and the training sample without turnover intention employee.
Specifically, the present embodiment is by gathering SMS historical record and/or the instant messaging historical record of training sample, it is thus achieved that
The user behavior data corresponding with history chat data entry of training sample.And assume the present embodiment gather except history chat number
User behavior data corresponding to other 5 leaving office attributes entries outside according to entry is the most as shown in table 1.
Step S403, use word frequency against text algorithm obtain the feature of the user behavior data corresponding with history chat data entry to
Amount.Specifically, it is assumed that the leaving office Feature Words list that the present embodiment pre-sets for " changing jobs ", " job hunting ", " recruitment ", " look for
Work ", " leaving office ", " resignation ", and reference simplifies in embodiment one the user row corresponding with history chat data entry of acquisition
For data (history chat text Tj) characteristic vector be W={w (t1,Tj)、w(t2,Tj)、w(t3,Tj)、w(t4,Tj)、
w(t5,Tj)、w(t6,Tj)}。
Step S404, according to predefined mark rule, to other leaving office attributes entries pair in addition to history chat data entry
The user behavior data answered carries out quantitative identifying, it is thus achieved that the characteristic vector of the user behavior data that other leaving office attributes entries are corresponding.
Specifically, with reference to simplifying user's row that other leaving office attributes entries obtained in embodiment two in addition to history chat data entry are corresponding
Method for the characteristic vector of data, it is assumed that the present embodiment get with other five leaving office attributes entries (job performance entry,
Job tenure entry, income level entry, working distance entry, log in recruitment and job hunting net frequency entries) corresponding user behavior
The ident value of data is respectively { 50,95,65,59,70}.
Step S405, according to characteristic vector and other leaving office attributes of the user behavior data corresponding with history chat data entry
The characteristic vector of the user behavior data that entry is corresponding, it is thus achieved that the characteristic vector of training sample.
According to step S403, the present embodiment for the characteristic vector of the user behavior data that history chat data entry obtains is
W={w (t1,Tj)、w(t2,Tj)、w(t3,Tj)、w(t4,Tj)、w(t5,Tj)、w(t6,Tj), and for job performance entry,
The user behavior number that job tenure entry, income level entry, working distance entry, login recruitment and job hunting net frequency entries obtain
According to characteristic vector be respectively { 50}, { 95}, { 65}, { 59}, { 70}.Therefore the method that the present embodiment uses " 0 " to fill will
For job performance entry, job tenure entry, income level entry, working distance entry, log in recruitment and job hunting net frequency bar
The dimension of the characteristic vector of the user behavior data that mesh obtains extends to the user behavior obtained for history chat data entry respectively
The dimension of the characteristic vector of data, namely will be less than the characteristic vector of 6 DOF, the method all using " 0 " to fill is extended to six
Dimension, thus the characteristic vector that finally can obtain training sample is 6*6 dimension.
Step S406, trains grader according to characteristic vector, it is thus achieved that leaving office forecast model.Specifically, it is assumed that the instruction of the present embodiment
Practicing total sample number is N, the most respectively the characteristic vector of each training sample is inputted grader and is trained, thus obtains leaving office prediction
Model, it should be noted that in order to obtain of a relatively high classification accuracy and precision of prediction, the training sample that the present embodiment is chosen
Quantity should be tried one's best greatly.
Step S407, according to leaving office forecast model, determines whether employee to be predicted has turnover intention.
The method of the prediction labor turnover of the present embodiment, by gathering the corresponding with leaving office attributes entries set in advance of training sample
User behavior data, and based on obtain user behavior data extract training sample characteristic vector, and based on extract spy
Property vector training for predicting whether employee to be predicted has the leaving office forecast model of turnover intention, solve and how to predict labor turnover
Technical problem, it is achieved that just turnover intention whether can be had to be predicted it according to the user behavior data of employee to be predicted, have
It is beneficial to enterprise and knows whether employee has turnover intention early, and take corresponding measure to reduce enterprise's separation rate, thus be greatly saved
Enterprise recruits spent manpower or monetary cost and the normal operation having ensured enterprise or work progress again.
The foregoing is only the preferred embodiments of the present invention, be not limited to the present invention, those skilled in the art is come
Saying, the present invention can have various modifications and variations.All within the spirit and principles in the present invention, any amendment of being made, equivalent
Replacement, improvement etc., should be included within the scope of the present invention.
Claims (7)
1. the method predicting labor turnover, it is characterised in that including:
Preset leaving office attributes entries;
Gathering the user behavior data corresponding with described leaving office attributes entries of training sample, wherein, described training sample includes turnover intention employee and the training sample without turnover intention employee;
Based on described user behavior data, extract the characteristic vector of described training sample;
Grader is trained, it is thus achieved that leaving office forecast model according to described characteristic vector;
According to described leaving office forecast model, determine whether employee to be predicted has turnover intention.
The method of prediction labor turnover the most according to claim 1, it is characterised in that described leaving office attributes entries includes:
History chat data, job performance, job tenure, income level, the last time promote time interval, working distance, one or more combinations of logging in recruitment and job hunting net frequency entries.
The method of prediction labor turnover the most according to claim 2, it is characterised in that the user behavior data corresponding with described history chat data entry gathering training sample includes:
Gather SMS historical record and/or the instant messaging historical record of training sample, as the user behavior data corresponding with described history chat data entry of training sample.
The method of prediction labor turnover the most according to claim 3, it is characterised in that based on described user behavior data, the characteristic vector extracting described training sample includes:
Word frequency is used to obtain the characteristic vector of the user behavior data corresponding with described history chat data entry against text algorithm;
According to predefined mark rule, the user behavior data that other leaving office attributes entries in addition to described history chat data entry are corresponding is carried out quantitative identifying, it is thus achieved that the characteristic vector of the user behavior data that other leaving office attributes entries are corresponding;
Characteristic vector according to the user behavior data corresponding with described history chat data entry and the characteristic vector of user behavior data corresponding to other leaving office attributes entries, it is thus achieved that the characteristic vector of described training sample.
The method of prediction labor turnover the most according to claim 4, it is characterised in that the characteristic vector using word frequency to obtain the user behavior data corresponding with described history chat data entry against text algorithm includes:
The user behavior data corresponding with described history chat data entry is converted into the character string of text formatting, it is thus achieved that history chat text;
Described history chat text is carried out participle, semantic disambiguation, removes stop words operation, it is thus achieved that participle text;
Word frequency is used to obtain the weighted value of the participle text mated in described participle text with default leaving office Feature Words against text algorithm, and using described weighted value as the characteristic vector of the user behavior data corresponding with described history chat data entry.
The method of prediction labor turnover the most according to claim 5, it is characterised in that according to described leaving office forecast model, determines whether employee to be predicted has turnover intention to include:
Gather the to be predicted user behavior data corresponding with described leaving office attributes entries of employee to be predicted;
Based on described user behavior data to be predicted, extract the characteristic vector of described user behavior data to be predicted;
Characteristic vector according to described user behavior data to be predicted and described leaving office forecast model, determine whether described employee to be predicted has turnover intention.
The method of prediction labor turnover the most according to claim 6, it is characterised in that described grader includes:
Any one in support vector machine classifier, Bayes classifier, maximum entropy classifiers.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610300752.9A CN106022708A (en) | 2016-05-09 | 2016-05-09 | Method for predicting employee resignation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610300752.9A CN106022708A (en) | 2016-05-09 | 2016-05-09 | Method for predicting employee resignation |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106022708A true CN106022708A (en) | 2016-10-12 |
Family
ID=57098802
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610300752.9A Pending CN106022708A (en) | 2016-05-09 | 2016-05-09 | Method for predicting employee resignation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106022708A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106529772A (en) * | 2016-10-31 | 2017-03-22 | 长沙军鸽软件有限公司 | Method of automatically acquiring team construction scheme |
CN107993085A (en) * | 2017-10-19 | 2018-05-04 | 阿里巴巴集团控股有限公司 | Model training method, the user's behavior prediction method and device based on model |
CN108337290A (en) * | 2017-12-26 | 2018-07-27 | 努比亚技术有限公司 | A kind of method and apparatus of enterprise staff user's behaviors analysis |
CN108427758A (en) * | 2018-03-19 | 2018-08-21 | 深信服科技股份有限公司 | A kind of leaving office trend analysis method, apparatus, equipment and storage medium |
CN108932525A (en) * | 2018-06-07 | 2018-12-04 | 阿里巴巴集团控股有限公司 | A kind of behavior prediction method and device |
CN109635092A (en) * | 2018-12-14 | 2019-04-16 | 泰康保险集团股份有限公司 | Analyze method, apparatus, medium and the electronic equipment of employee's working condition |
WO2020019489A1 (en) * | 2018-07-25 | 2020-01-30 | 平安科技(深圳)有限公司 | Method for predicting reason for employee resignation and related device |
CN111639814A (en) * | 2020-06-02 | 2020-09-08 | 贝壳技术有限公司 | Method, apparatus, medium, and electronic device for predicting occurrence probability of fluctuating behavior |
CN111798059A (en) * | 2020-07-10 | 2020-10-20 | 河北冀联人力资源服务集团有限公司 | System and method for predicting job leaving |
CN112257777A (en) * | 2020-10-21 | 2021-01-22 | 平安科技(深圳)有限公司 | Off-job prediction method based on hidden Markov model and related device |
CN112381444A (en) * | 2020-11-26 | 2021-02-19 | 方是哲如管理咨询有限公司 | Organization risk prediction method and system based on organization member behavior data |
CN112488437A (en) * | 2019-09-12 | 2021-03-12 | 英业达科技有限公司 | Human resource management system and method thereof |
CN113762621A (en) * | 2021-09-09 | 2021-12-07 | 南京领行科技股份有限公司 | Network taxi appointment driver departure prediction method and system |
TWI777154B (en) * | 2020-04-23 | 2022-09-11 | 和碩聯合科技股份有限公司 | Electronic device and method for turnover rate prediction |
CN117408660A (en) * | 2023-12-15 | 2024-01-16 | 山东杰出人才发展集团有限公司 | Human resource data service management system based on big data |
-
2016
- 2016-05-09 CN CN201610300752.9A patent/CN106022708A/en active Pending
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106529772A (en) * | 2016-10-31 | 2017-03-22 | 长沙军鸽软件有限公司 | Method of automatically acquiring team construction scheme |
CN107993085B (en) * | 2017-10-19 | 2021-05-18 | 创新先进技术有限公司 | Model training method, and user behavior prediction method and device based on model |
CN107993085A (en) * | 2017-10-19 | 2018-05-04 | 阿里巴巴集团控股有限公司 | Model training method, the user's behavior prediction method and device based on model |
CN108337290A (en) * | 2017-12-26 | 2018-07-27 | 努比亚技术有限公司 | A kind of method and apparatus of enterprise staff user's behaviors analysis |
CN108427758A (en) * | 2018-03-19 | 2018-08-21 | 深信服科技股份有限公司 | A kind of leaving office trend analysis method, apparatus, equipment and storage medium |
CN108932525A (en) * | 2018-06-07 | 2018-12-04 | 阿里巴巴集团控股有限公司 | A kind of behavior prediction method and device |
WO2020019489A1 (en) * | 2018-07-25 | 2020-01-30 | 平安科技(深圳)有限公司 | Method for predicting reason for employee resignation and related device |
CN109635092A (en) * | 2018-12-14 | 2019-04-16 | 泰康保险集团股份有限公司 | Analyze method, apparatus, medium and the electronic equipment of employee's working condition |
CN112488437A (en) * | 2019-09-12 | 2021-03-12 | 英业达科技有限公司 | Human resource management system and method thereof |
TWI777154B (en) * | 2020-04-23 | 2022-09-11 | 和碩聯合科技股份有限公司 | Electronic device and method for turnover rate prediction |
CN111639814A (en) * | 2020-06-02 | 2020-09-08 | 贝壳技术有限公司 | Method, apparatus, medium, and electronic device for predicting occurrence probability of fluctuating behavior |
CN111798059A (en) * | 2020-07-10 | 2020-10-20 | 河北冀联人力资源服务集团有限公司 | System and method for predicting job leaving |
CN111798059B (en) * | 2020-07-10 | 2023-11-24 | 河北冀联人力资源服务集团有限公司 | Off-duty prediction system and method |
CN112257777A (en) * | 2020-10-21 | 2021-01-22 | 平安科技(深圳)有限公司 | Off-job prediction method based on hidden Markov model and related device |
WO2021179715A1 (en) * | 2020-10-21 | 2021-09-16 | 平安科技(深圳)有限公司 | Hidden markov model-based resignation prediction method and related device |
CN112257777B (en) * | 2020-10-21 | 2023-09-05 | 平安科技(深圳)有限公司 | Off-duty prediction method and related device based on hidden Markov model |
CN112381444A (en) * | 2020-11-26 | 2021-02-19 | 方是哲如管理咨询有限公司 | Organization risk prediction method and system based on organization member behavior data |
CN113762621A (en) * | 2021-09-09 | 2021-12-07 | 南京领行科技股份有限公司 | Network taxi appointment driver departure prediction method and system |
CN117408660A (en) * | 2023-12-15 | 2024-01-16 | 山东杰出人才发展集团有限公司 | Human resource data service management system based on big data |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106022708A (en) | Method for predicting employee resignation | |
CN110472017A (en) | A kind of analysis of words art and topic point identify matched method and system | |
CN108052605A (en) | A kind of intelligent Answer System based on client feature library | |
CN109902216A (en) | A kind of data collection and analysis method based on social networks | |
CN106357942A (en) | Intelligent response method and system based on context dialogue semantic recognition | |
CN106021389A (en) | System and method for automatically generating news based on template | |
CN105894253A (en) | Method and device for automatic pushing of job application demand | |
CN111182162B (en) | Telephone quality inspection method, device, equipment and storage medium based on artificial intelligence | |
CN103593412B (en) | A kind of answer method and system based on tree structure problem | |
CN110516057B (en) | Petition question answering method and device | |
CN106886567A (en) | Microblogging incident detection method and device based on semantic extension | |
CN105868183A (en) | Method and device for predicting staff demission | |
CN105808721A (en) | Data mining based customer service content analysis method and system | |
CN110334214A (en) | A kind of method of false lawsuit in automatic identification case | |
CN107784033A (en) | A kind of dialogue-based method and apparatus recommended | |
CN110880142B (en) | Risk entity acquisition method and device | |
CN109947934A (en) | For the data digging method and system of short text | |
CN111061837A (en) | Topic identification method, device, equipment and medium | |
CN111782611B (en) | Prediction model modeling method, device, equipment and storage medium | |
CN107391581A (en) | Community network information dissemination Forecasting Methodology and equipment | |
CN109284500A (en) | Information transmission system and method based on merchants inviting work process and reading preference | |
CN116049379A (en) | Knowledge recommendation method, knowledge recommendation device, electronic equipment and storage medium | |
CN107679209A (en) | Expression formula generation method of classifying and device | |
CN113379432B (en) | Sales system customer matching method based on machine learning | |
CN110413307A (en) | Correlating method, device and the electronic equipment of code function |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20161012 |
|
RJ01 | Rejection of invention patent application after publication |