Subscribe to our Newsletter


UKrecruiter Newsletter

Hi everyone,

I hope you all enjoyed the Jubilee Weekend and are refreshed and raring to go with your Internet Recruitment strategy.

We have looked at a number of different recruitment solutions (with more to come). Most have one thing in common – simple text search as the means of matching candidates to jobs and vice versa. (The exception was Zynap).

How do we as recruiters (or line managers) actually ‘sift’ CVs and match them to job requirements? We scan the CV visually, highlighting key information – location, education, skills, employment history, current employer, salary and other specialist information that may be relevant to our company and jobs. Visually we might weight a certain skill higher if it occurs in actual recent work experience as opposed to being in a course or even out of date. We then make the “yes, no, maybe” decision and ‘file’ the CV in the appropriate channel for action.

What does simple text search mean in practice? We try and create a search string that will replicate the above process. This can be complicated, as there are a number of variables. Typically, the search, if too loosely defined, brings back too many candidates and if too tightly defined, too few responses. Why does this happen?

A simple keyword matching technology allows you to identify only things that are stored in a list of known keywords. An advancement has been the addition of pattern matching technology where the system can recognize things stored in the lists allowing for some variations. (Many vendors tried to sell this as ‘fuzzy logic’)

Another challenge is that different contexts impose different meanings to words and phrases. For instance, “Project Manager” is a potential skill for a candidate but in the sentence “I was reporting to the Project Manager”, it is not. Birmingham may be where the candidate resides, where the place of work is or where the degree was earned.

A further complication is that the terminology used in the job requisition may not actually match the language usage that the candidate is more comfortable with. If you run an ad for a Bin Man or Refuse Collector and I call myself a Sanitation Engineer, the simple text search will not provide a match.

Hence, there are two key issues not addressed by this technology:

  • · How to construct concise, accurate and uniform profiles from CVs and job advertisements automatically
  • · How to find relevant candidates/vacancies whose profiles are not directly matched by a particular search criteria or saved search string.


So, what to do? There have been a number of attempts to utilise Artificial Intelligence to address this issue. Resumix was the first and most successful. Zynap is getting there. The newest player in this market is Infogistics Limited, a leading provider of text-analysis solutions across multiple markets including HR, law enforcement, knowledge management and CRM. It is an Edinburgh-based company, launched in July 2000 as a spin-off from the University of Edinburgh and founded by internationally recognized experts in the fields of text-mining and document retrieval.

Infogistics has launched a set of intelligent HR tools for extracting information from the CV (CVXtractor – automatic profile extraction engine) and searching candidate databases, either your own or external sites. (RealTerm – candidate search and browsing engine). The major benefits of Infogistics’ intelligent HR tools are:

  • · Unique integrated search and browsing functionality that enables flexible mix-and-match strategy in linking candidates to vacancies · Real-time automatic population of candidate and vacancy databases from multiple on-line sources
  • · Labour and time savings on candidate and vacancy profile construction
  • · Guaranteed consistency between profiles created by different recruiters or candidates


CVXtractor imitates the way humans read and understand: it analyses the linguistics structure of a text passage and applies context-driven reasoning to figure out what different words mean in a particular context and how they can be grouped together. It can identify work experience facts including time period, employer and position held. Similarly it can identify education and qualification facts including degree/qualification, school and the date achieved. And, of course, it can identify industry specific skills to a high degree of accuracy. Infogistics also supplies a very efficient post-editing interface, which shows the extracted facts and the exact locations in the CV where they were found. CVXtractor, has demonstrably speeded up profile construction by up to 10 times over standard manual methods. (see the end of the article for the “How does it do this?” detailed explanation)

RealTerm is a topic identification and clustering technology. It analyses results returned by a search, organizes these results into a hierarchy of topics and presents them to you for browsing and exploring. RealTerm enables a highly intuitive candidate-vacancy matching process. Like a traditional search, RealTerm lets you formulate a query and submit it to the system. It then retrieves candidate profiles with the required skills and qualifications and ranks them according to how many of the target criteria have matched information in the profiles. Unique to RealTerm, a list of secondary skills, qualifications and preferred locations is also presented to the user. Using this additional information the user can quickly and intuitively browse through the set of returned candidates’ profiles and populate the target hot list in a matter of seconds.

The Infogistics software can be integrated easily into existing workflow systems, adapted to most of the existing candidate management and HRIS systems or be used on an ad hoc basis to search external databases.

Xtractor – automatic profile extraction engine Infogistics technology is based on recent advances in Artificial Intelligence and Computational Linguistics. First it partitions a resume into zones including personal information, overall summary, employment history, and education background. Then it treats information extracted from these different zones differently: a programming language mentioned in the employment history is a much stronger skill than one mentioned in the education background, a town mentioned in the address of a previous employer is not necessarily the location where the candidate resides or seeks a job. Extracted qualities are then normalized to their main forms. For instance, the word “Novell” is normalized to “Novell Netware” if it stands for a software product or “Novell Inc.” if it stands for the company itself.

Comments are closed.