WO2017035556A1

WO2017035556A1 - A candidate identification system

Info

Publication number: WO2017035556A1
Application number: PCT/AU2015/000653
Authority: WO
Inventors: Eugene SVISTUNOV; Antony UGONI
Original assignee: Seek Ltd
Current assignee: Seek Ltd
Priority date: 2015-10-30
Filing date: 2015-10-30
Publication date: 2017-03-09
Anticipated expiration: 2018-04-30

Abstract

A method to identify candidates for jobs which involves processing interaction data of candidates to derive for each candidate, time interval data for a plurality of time periods, the time interval data indicative of candidate behaviour, the interaction data defining interactions between the candidate and a job advertising platform; processing candidate profiles to derive profile data for each candidate; inputting the time interval data and profile data of the candidates to a propensity model derived from prior candidate interactions with the platform; receiving from the propensity model, propensity data for each candidate indicative of the candidate's propensity to apply for a job within a decision period; receiving a hirer query specifying characteristics of a desired candidate; generating a candidate list by matching the desired candidate characteristics to candidate profiles; outputting the candidate list in conjunction with propensity data for at least one candidate of the candidate list.

Description

TITLE

A CANDIDATE IDENTIFICATION SYSTEM FIELD

The present invention relates to an electronic method and a candidate identification system for assisting hirers to identify candidates for jobs.

BACKGROUND

In recent years the advertising of jobs has trended away from traditional advertising in newspapers to online advertisement of jobs via websites. Over time, such websites have developed a database of candidates who are either currently looking for employment or have previously used the job advertising website. In recent times, such websites have added the facility for potential hirers to search their candidate databases to locate candidates that match their requirements without needing candidates to reply to a job advertisement. The search engines employed by such websites typically match criteria input by the hirer against candidate profiles which describe details of the type of job the candidate is seeking or the role that they currently hold.

A problem with such arrangements is that the search results will inevitably include a number of people who may not be open to an approach for employment. Accordingly, current systems require the hirer to dedicate significant energy to sorting through the potential candidates to locate candidates who are interested in a specific job. There is a need for an improved method for assisting hirers to identify candidates . SUMMARY OF INVENTION

In a first aspect, the invention provides an electronic method of assisting hirers to identify candidates for jobs, the method comprising:

hosting, on a web server, a job advertising platform on which a hirer can place job advertisements and via which a candidate can establish a candidate profile in a candidate datastore ;

monitoring interactions between candidates and the job advertising platform and storing interaction data that defines interactions between the respective candidate and the job advertising web site;

processing the interaction data of a plurality of candidates to derive for each respective candidate, time interval data for a plurality of time periods , the time interval data of each time period indicative of candidate behaviour in the respective time period;

processing candidate profiles of the plurality of candidates from the candidate datastore, to derive profile data for each of the plurality of candidates ;

inputting at least the time interval data and profile data of the plurality of candidates to a propensity model derived, at least in part, from prior candidate

interactions with the job advertising platform;

receiving as an output of the propensity model, propensity data for each candidate, the propensity data indicative of the respective candidates propensity to apply for a job within a decision period;

receiving a hirer query specifying characteristics of a hirer job;

generating, with a search engine, a candidate list from the plurality of candidates by matching the hirer job characteristics to candidate profiles of the plurality of candidates ; outputting, to the hirer, the candidate list in conjunction with propensity data for at least one

candidate of the candidate list. Thus , hirers are provided with an indication as to whether a prospective candidate is likely to be receptive to an approach for employment.

In an embodiment, the interaction data is indicative of characteristics of the job and when the job advertisement was viewed .

In an embodiment, processing of the interaction data is arranged so that the time interval data is indicative of a rate of interaction with job advertisements.

In an embodiment, processing of the interaction data is arranged so that the time interval data is indicative of shared characteristics of job advertisements interacted with during the time interval.

In an embodiment, the time intervals are a series of contiguous time intervals preceding a defined time

relative to which the interaction data is processed.

In an embodiment, the time intervals are of varying length .

In an embodiment, the time intervals closer to the defined time are shorter than the time intervals further from the defined time.

In a second aspect, the invention provides, a candidate identification system for assisting hirers to identify candidates for jobs, the candidate identification system comprising: a web server hosting a job advertising platform on which a hirer can place job advertisements and via which a candidate can establish a candidate profile in a candidate datastore ;

an interaction monitor arranged to monitor

interactions between candidates and the job advertising platform and store interaction data that defines

interactions between the respective candidate and the job advertising web site;

a data conditioner arranged to

a) process the interaction data of a plurality of candidates to derive for each respective candidate, time interval data for a plurality of time periods , the time interval data of each time period indicative of candidate behaviour the respective time period, and

b) process candidate profiles of the plurality of candidates from the candidate datastore, to derive profile data for each of the plurality of

candidates ;

a propensity model derived, at least in part, from prior candidate interactions with the job advertising platform, the propensity model arranged to receive at least the time interval data and profile data of the plurality of candidates and output propensity data for each candidate, the propensity data indicative of the respective candidates propensity to apply for a job within a decision period; and

a candidate search engine arranged to

a) process a hirer query specifying characteristics of a hirer job generate a candidate list from the plurality of candidates by matching the hirer job characteristics to candidate profiles of the plurality of candidates , and

b) output the candidate list in

conjunction with propensity data for at least one

candidate of the candidate list. In a third aspect, the invention provides computer program code which, when executed, implements the above method. BRIEF DESCRIPTION OF THE DRAWINGS

Embodiments of the invention will now be described in relation to the following drawings in which: Figure 1 is a block diagram of a candidate identification system of an embodiment of the invention; and

Figure 2 illustrates an example of propensity data in the form of an approachability signal .

DETAILED DESCRIPTION

In embodiments of the invention, the candidate

identification system 100 generates propensity data in the form of an approachability signal using a propensity model. In an embodiment, the propensity model estimates each candidate' s propensity to apply for a job within the next a decision period. In one example, the decision period may be 7 days .

The approachability signal indicates a candidate' s likelihood to apply for a job - their approachability. In an example, the approachability signal is available via a hirer web portal (a candidate datastore available to advertisers on the site) .

In embodiments of the invention, the approachability signal is output as a message 210 to the hirer in a list of candidates delivered via a web page as shown in Figure 2 . An example message is: "candidates like this tend to be approachable". In an embodiment, a hover box 220 is provided so that when a user places a pointing device such as a mouse (or in touch screen embodiments, a finger) over the message, an additional message is provided that explains the basis for the approachability signal, for example "this is based on similar profiles of candidates who have been open to discussing career opportunities (e.g. similar industry, tenure, online activity etc.)".

As shown schematically in Figure 1 , candidates and hirers interact with candidate identification system 100 over a network 50, such as the internet 50, using web browsers or applications executing on candidate devices 10 and hirer devices 20 respectively. That is, candidates and hirers interface with the job advertising platform 110 either via a web site or via an application. In one example, there may be separate applications for candidates and hirers. To this end, both the job advertising platform 110 with which the candidates interact and a hirer portal 120 are hosted by a web server 105. The hirer portal 120 (or a hirer application) can be used to add advertisements to the job advertising platform 110 or to conduct searches for candidates . Persons skilled in the art will appreciate that, as is known in the art, the web server may be split across a number of physical computing devices in order to manage issues such as load balancing.

The job advertising platform 110 is configured to allow candidates to search for jobs advertised using a job search engine (not shown) . The job advertising platform 110 also allows users to establish a profile within candidate datastore 141 of system datastore 140.

Candidates 10 register with the system using a

conventional registration process by which various characteristics of the candidate are captured. The registration process also results in a candidate

identifier being assigned to each candidate. Similarly, when the hirer portal 120 is used to add jobs to the job advertising platform these are stored in a job datastore 143, a job advertisement ID is assigned, and

characteristics of the jobs are included therein.

In an embodiment, the system 100 includes an interaction monitor 130 which monitors for specific interactions between candidates 10 and the job advertising platform, and updates the interactions events 142 stored in system datastore 140 to reflect these interactions as described in further detail below.

System datastore 140 is a relational datastore hence it will be appreciated that as a candidate identifier (ID) is captured for each interaction event 142, each interaction event generated in respect of an interaction by a

particular candidate will be stored in the system

datastore 140 in association with the respective

candidate. While the system datastore 140 is shown as having separate candidate datastore 141, interaction event datastore 142 and job advertisement datastore 143, however this is only an example of an arrangement of the data in a set of data structures within the relational datastore. That is , due to the relational nature of the system datastore 140, while certain information is stored in particular data structures within the datastore 140, the data is linked by relationships.

The system includes a data conditioner 150 for preprocessing the interaction events as well as aspects of the candidate profiles before these variables are fed to the propensity model 160. As explained in further detail below, the propensity model 160 outputs the

approachability signal described above for each of the candidates in the candidate datastore. This is stored as propensity data 144 in association with the respective candidate profiles in candidate datastore 141 so that when the hirer inputs a query to search engine 170, it can be retrieved along with the candidate profile. In embodiments of the invention, propensity data is

calculated for all candidates on a daily basis so that it is available and up to date each time a query is input via search engine 170. However, it will be appreciated that in other embodiments , the propensity model may only calculate propensity data for a subset of candidates , for example, those that have been active during a particular time period. It will also be appreciated from the

description below, that where the data of all candidates are processed, there may not be values for all of the variables. In an embodiment in such cases preprocessing results in these variables as being marked as "missing values" and they are fed to the propensity model as part of the data associated with specific candidates and hence, the "missing values", are taken into account by the propensity model .

As indicated above, the interaction monitor 130 processes interactions events between logged-in candidates 10 whose candidate ID can be determined and the job advertising platform 110. In the embodiment, the interaction events processed by interaction monitor 130 in the embodiment are divided into three separate event types :

• job advertisement views by candidates;

· job advertisement applications by candidates; and

• candidates creating a favourite' job advertisement by saving the job advertisement.

Persons skilled in the art will appreciate that

alternative event types could be used in other

embodiments, for example, job searches instead of job views .

Each of these events is stored in the system datastore 140 as an interaction event 142. The interaction events 142 have the following common attributes which are saved and used by the propensity model when the propensity data is calculated:

• candidate ID ;

· Job Advertisement ID;

• Date & Time of the interaction; and

• Event Type of the three event types described

above .

The job advertisement ID allows data from the stored job advertisements 143 to be associated with specific

interaction events. By way of example, in one embodiment, the following job advertisement 'raw' attributes are used:

• Salary ;

• Salary Type (eg. Hourly rate, Annual, Commission, etc) ;

• Work type (eg. Full time, Contract, Part time,

Casual) ; and

• Advertisement Classification (eg . Real Estate & Property, Science & Technology, Trades & Services, etc) .

•

It will be appreciated that while in the embodiment, capturing the job advertisement ID enables this data to be retrieved when needed for processing, in other embodiments it could be captured by the interaction monitor 130 and stored as part of an interaction event rather than by capturing the job advertisement ID.

Similarly, the candidate ID enables candidate data to be retrieved when the data is processed. In the embodiment, the following Candidate profile raw variables are used:

• "'I'm available in XX weeks" data element

• Profile work history with employment dates

• Profile aspirational classification

Data conditioning As indicated above, some of the data captured in relation to the interaction events is processed by the data

conditioner 150 in order to be placed in an a condition suitable for presentation to the propensity model 160. In the embodiment, a number of variables are derived from the interaction events for a plurality of time intervals that precede the date at which the propensity data is

generated. In the embodiment, the time intervals are a set of contiguous time intervals measured in days . It will be apparent that other time intervals could be used.

By way of example, in the embodiment, the derived

variables are :

• number of job detail views within the time interval;

• number of applications within the time interval;

• number of jobs favorited (saved ) within the time

interval ;

• the average salary (annualised) of the jobs viewed

within the time interval ;

• the mode of the work type (eg. Full time, Contract, Part time, Casual) of the jobs viewed within the time

interval (i.e. the work type looked at most frequently); and

• the mode of the job classification (eg. Real Estate & Property, Science & Technology, Trades & Services, etc) viewed within the time interval .

All of these variables relate to behavior of the

candidate. In one example, the interaction events are calculated back from the current date (= interaction event date - the beginning of the decision period) nine times in variable length intervals of 1 , 2 , 3-4, 5-7, 8-12 , 13- 20, 21-33, 34-54 and 55-88 days ago (a Fibonacci

sequence) , which makes 54 features that are provided to the propensity model 160. It will be appreciated that it is generally desirable that the time intervals get longer the further back in time relative to the decision period in order to give more weight to recent interaction events .

In addition, the data conditioner 150 derives profile features from the candidate profile data (obtained via candidate , namely :

• number of distinct employers in the profile work

history; and

• tenure in the most recent (or current) role.

Propensity model

The propensity model is derived using the same variables derived from the interaction events 142 as described above but these variable derived at an earlier time period in a training mode of the propensity model 160.

In one example, the propensity model 160 is built as a binary classification model in Salford Predictive Modeller using Treenet. Treenet is a commercial product

implementing stochastic Gradient Boosted Models with classification trees (http : //www . salford- systems . com/products/treenet) . In one example, the model is trained on a sample of approx. 400,000 interaction events and tested/validated on a holdout set of approx. 80,000 interaction events.

In an embodiment, the propensity model 160 is built using stochastic gradient boosting, see

https : //en . wikipedia . org/wiki/Gradient boosting - and explained in details by the author at - http: //statweb. stanford.edu/~jhf/ftp/stobst.pdf .

The initial Treenet model fit used the following settings : 600 trees, tree size 6 terminal nodes, minimum 200

observations in the terminal node, learn rate 0.05, subsample rate 0.7. , standard influence trimming. The second stage Importance Sampled Learning Ensembles model was fit to 600 trees with the elasticity 1.1.

In an embodiment, the initial output of the propensity model is in the form of a raw score in the range of 0-100. In an embodiment, this score could be output as the propensity data to hirers. In an embodiment, rather than displaying the score, a threshold is applied to the scores and propensity data in the form of the availability signal describe above is displayed for candidates having a score above the threshold. In an embodiment, the threshold is a value in the range of 0-100 that corresponds to the candidate being three times more likely to apply from than the random probability of a candidate applying.

Candidate matching

A hirer 20 inputs a search request via the hirer portal 120 either by formulating a query or submitting a job advertisement. In an embodiment, the search engine 170 identifies key concepts (i.e. Role title, seniority, etc) within the job advertisement and matches that data

to associated fields within candidate profiles 141. In another example, the hirer specifies a desired candidate skill set. Thus, either via the action or the hirer or the search engine a set of desired candidate characteristics are obtained which can be used to retrieve a set of candidates from candidates 141. The search engine

then determines an overall rank score for each profile in the retrieval set, and the results are sorted by the rank score to produce a candidate list. If there is an

availability signal for a specific candidate, it is included in the candidate list. The candidate list is output to the hirer via the browser executing on hirer device 20.

Overview of method Figure 3 shows a summary of the method 300 of an

embodiment. The method involves monitoring 310

interactions with the job advertising platform to obtain interaction data. The interactions are processed 320 to get time interval data. Candidate profiles are also processed 330 to derive profile data. The time interval data and the profile data are input 340 to the propensity model which outputs 350 propensity data. The system receives 360 a candidate query from a hirer and outputs a list of candidates together with any propensity data 370 for those candidates .

Further aspects of the method will be apparent from the above description of the system. It will be appreciated that the method will be implemented electronically, for example, digitally by a processor executing program code. In this respect, in the above description certain steps are described as being carried out by a processor, it will be appreciated that such steps will often require a number of sub-steps to be carried out for the steps to be

implemented electronically, for example due to hardware or programming limitations. For example, to carry out a step such as matching a processor may need to compute several values and compare those values.

As indicated above, the method may be embodied in program code . The program code could be supplied in a number of ways, for example on a tangible computer readable storage medium, such as a disc or a memory device, e.g. an EEPROM, or as a data signal (for example, by transmitting it from a server) . Further different parts of the program code can be executed by different devices, for example in a client server relationship. Persons skilled in the art, will appreciate that program code provides a series of

instructions executable by the processor. Herein the term "processor" is used to refer generically to any device that can process instructions and may include: a microprocessor, a general purpose computer (e.g. a PC) or a server. That is a processor may be provided by any suitable logic circuitry for receiving inputs, processing them in accordance with instructions stored in memory and generating outputs (for example on the display) . Such processors are sometimes also referred to as central processing units (CPUs) . Most processors are general purpose units, however, it is also know to provide a specific purpose processor, for example, an application specific integrated circuit (ASIC) or a field programmable gate array (FPGA) . It will be understood to persons skilled in the art of the invention that many modifications may be made without departing from the spirit and scope of the invention, in particular it will be apparent that certain features of embodiments of the invention can be employed to form further embodiments.

It is to be understood that, if any prior art is referred to herein, such reference does not constitute an admission that the prior art forms a part of the common general knowledge in the art in any country.

In the claims which follow and in the preceding

description of the invention, except where the context requires otherwise due to express language or necessary implication, the word "comprise" or variations such as

"comprises" or "comprising" is used in an inclusive sense, i.e. to specify the presence of the stated features but not to preclude the presence or addition of further features in various embodiments of the invention.

Claims

CLAIMS :

1. An electronic method of assisting hirers to identify candidates for jobs, the method comprising: processing interaction data of a plurality of candidates to derive for each respective candidate, time interval data for a plurality of time periods , the time interval data of each time period indicative of candidate behaviour the respective time period, wherein the interaction data defines interactions between the respective candidate and a job

advertising platform on which job advertisements are placed by hirers ;

processing candidate profiles of the plurality of candidates from a candidate datastore, to derive profile data for each of the plurality of candidates ; inputting at least the time interval data and profile data of the plurality of candidates to a propensity model derived, at least in part, from prior candidate interactions with the job advertising platform;

receiving as an output of the propensity model, propensity data for each candidate, the propensity data indicative of the respective candidates

propensity to apply for a job within a decision period;

receiving a hirer query specifying

characteristics of a desired candidate;

generating, with a search engine, a candidate list from the plurality of candidates by matching the desired candidate characteristics to candidate profiles of the plurality of candidates;

outputting, to the hirer, the candidate list in conjunction with propensity data for at least one candidate of the candidate list.

2. A method as claimed in claim 1 , wherein

the interaction data is indicative of characteristics of the job and when the job advertisement was viewed.

3. A method as claimed in claim 1 or claim 2 , wherein processing of the interaction data is

arranged so that the time interval data is indicative of a rate of interaction with job advertisements.

4. A method as claimed in any one of claims 1 to 3 , wherein processing of the interaction data is

arranged so that the time interval data is indicative of shared characteristics of job advertisements interacted with during the time interval .

5. A method as claimed in any one of claims 1 to 4 wherein the time intervals are a series of contiguous time intervals preceding a defined time relative to which the interaction data is processed.

6. A method as claimed in claim 5 , wherein the time intervals are of varying length.

7. A method as claimed in claim 6 , wherein the time intervals closer to the defined time are shorter than the time intervals further from the defined time.

8. A method as claimed in any one of claims 1 to comprising hosting the job advertising platform on web server;

receiving candidate profiles via the job

advertising platform; and

monitoring interactions between candidates and the job advertising platform to generate the

interaction data.

9. A candidate identification system for assisting hirers to identify candidates for jobs, the candidate identification system comprising:

a data conditioner arranged to:

a) process interaction data of a plurality of candidates to derive for each respective

candidate, time interval data for a plurality of time periods , the time interval data of each time period indicative of candidate behaviour the respective time period, wherein the interaction data defines

interactions between the respective candidate and a job advertising platform on which job advertisements are placed by hirers , and

b) process candidate profiles of the plurality of candidates from a candidate datastore, to derive profile data for each of the plurality of candidates ;

a propensity model derived, at least in part, from prior candidate interactions with the job advertising platform, the propensity model arranged to receive at least the time interval data and profile data of the plurality of candidates and output propensity data for each candidate, the propensity data indicative of the respective

candidates propensity to apply for a job within a decision period; and

a candidate search engine arranged to

a) process a hirer query specifying desired candidate characteristics to generate a candidate list from the plurality of candidates by matching the desired candidate characteristics to candidate profiles of the plurality of candidates , and

b) output the candidate list in conjunction with propensity data for at least one candidate of the candidate list.

10 . A candidate identification system as claimed in claim 9, further comprising a web server hosting the job advertising platform via which a candidate can establish a candidate profile in a candidate

datastore and on which a hirer can place job

advertisements ; and

an interaction monitor arranged to monitor interactions between candidates and the job

advertising platform and store interaction data that defines interactions between the respective candidate and the job advertising web site.

11. A candidate identification system as claimed in claim 9 or claim 10, wherein the interaction data is indicative of characteristics of the job and when the job advertisement was viewed.

12. A candidate identification system as claimed in any one of claims 9 to 11, wherein the data

conditioner processes the interaction data so that the time interval data is indicative of a rate of interaction with job advertisements.

13. A candidate identification system as claimed in any one of claims 9 to 12 , wherein the data

conditioner processes the interaction data so that the time interval data is indicative of shared characteristics of job advertisements interacted with during the time interval . 1 . A candidate identification system as claimed in any one of claims 9 to 13 wherein the time intervals are a series of contiguous time intervals preceding a defined time relative to which the interaction data is processed.

15. A candidate identification system as claimed in claim 14, wherein the time intervals are of varying length .

16. A candidate identification system as claimed in claim 6, wherein the time intervals closer to the defined time are shorter than the time intervals further from the defined time.

17. Computer program code which when executed implements the method of any one of claims 1 to 8.

18. A tangible computer readable medium comprising the computer program code of claim 17.