Pattern-Based Matching
Which skills, work experiences, certifications, and educational credentials does this job seeker possess? How has she moved through her career and where is she likely to go next? What is this job fundamentally about and what kinds of prior experiences and skills are most important?
Burning Glass deploys the only technology on the market that is capable of answering those questions and then using the resulting information to match jobs seekers to jobs.
Using our patented parsing engine, we begin by reading and analyzing resumes and job descriptions in order to extract all information related to education, experience, knowledge, and skills, as well as career path. The system performs extensive conversion, interpretation, and normalization of extracted text so that it can be treated as quantitative data, and is also able to derive and infer as well as extract data. Once data for a given job seeker and job have been compiled, our Predictive Matching technology kicks in across two dimensions:
- By applying mathematical models of employer and job seeker behavior that determine the degree of compatibility between a job seeker and a job posting, our technology seeks out contextual correlations, both within a document and across documents, in order to identify conceptual similarity – not just matching of keywords.
- Based on millions of observed transitions, Burning Glass has developed a KnowledgeMine™ which characterizes entities (i.e. employers, educational institutions, job titles, degrees, etc.) and the long-term impact they have on careers. When assessing how likely it is for a job seeker to match a given job, our system refers to the KnowledgeMine for a probabilistic assessment of the potential transition.
The information is then weighted appropriately by a neural network model, resulting in a score that represents the mathematical probability of a match between the job seeker and job opening.
How It Works
Our Predictive Matching technology brings together two areas of machine learning that have not been combined in the past:
- Predictive modeling provides a powerful and general framework for forecasts or predictions, based on data that’s available to understand the individual or entity of interest. Relevant data are collected and converted into a characterization (profile) of the individual or entity, and a model is then constructed based on patterns learned from historical data which enables predictions or estimations of future outcomes. In the employment context, predictive modeling can be used to assess the compatibility of job seekers and job openings.
- Predictive modeling has traditionally been applied in settings where the underlying data is numerical. In the employment context, however, most of the relevant data comes in the form of resumes and job postings, which are typically free or semi-structured text documents. Statistical Natural Language Processing (SNLP) techniques enable extraction of relevant information from text documents for use in predictive models. SNLP also can be used to associate meaning with fragments of free text in order to assess similarity based on meaning rather than overlapping words or synonyms.
