I have written about semantic technology and its providers a good bit on this and other blogs (see for example: Is Semantic Technology Real?) so I was interested to see how Monster.com was applying it.   I spoke with Javid Muhammedali, their Director of Product Management on their efforts.

Javid started with some of the problems with key word search in the recruitment field. All words are given equal weight, you need complex Boolean logic to narrow down results, the search engine cannot distinguish between recent and dated information,,the search engine does not understand the meaning of words or their context, and you often have to sift through irrelevant results. This makes sense and I see it all the time in many contexts.

The literal nature of matches makes it necessary to look for many variations on a way to express something. Javid showed a three-part search that would need over 10,000 combinations to be fully covered.

Javid explained two ways that semantic search can help. First, it can understand the meaning of words and use the context to help determine meaning. Second, it understands the hierarchy of concepts in such areas as job titles, skills, industries, etc. The context allows it to understand the different uses of a word like, Washington, when used in an address, school, company, person, etc. This context comes from the immediately connected words (e.g., Ave., Mutual, University, and the location within a document (e.g., contact information, education).

The Monster semantic search engine can be used to search documents that are uploaded, as well as resumes created with their resume builder. The search engine is not taught literal rules but rather assigns weights to the various aspects being searched on based on past experience and customer feedback. As part of the process of fine tuning the search engine, they assign resumes to both the tool and to a panel of seasoned recruiters and made comparisons. Adjustments were made to the tool to better aligned with the expert human logic. For example, rare, but useful, skills were identified and given greater weighting than common skills.

Javid did some sample candidate searches. The results are rank ordered with the rationale clearly provided. It then distills the key points and allows you to compare candidates on these points. It also understands the connections between filters such as roles and companies and makes adjustments.

The goal is to narrow down a list of thousands or even hundreds of possible candidates to a few dozen that a person that easily sort through and make the final judgment on. The comparison feature comes in very handy once you get to this stage.  While we looked at the candidate results for those with jobs, the same approach is available for job seekers.

The results have been promising so far. Monster has found a 65% time savings using the semantic Power Resume Search compared to keyword-based search engines. There has also been a 150% increase in the number of qualified candidates found.  In this application, semantic search seems both real and useful.

Monster is also attempting to have more user input through social media type features. You can rate candidates as a user such like many travel sites. Javid said that the take-up on this has been slow. A challenge for him in the coming year is to figure out how to increase the engagement with these ratings. This is a useful goal and I would suggest looking at what other successful social media efforts have done.