How does ALEX process skills?
ALEX employs several AI parsing strategies including natural language processing techniques in order to extract relevant information from resumes written in a free-text format. The use of several techniques in combination allows ALEX to try and interpret the same data several times if needed by applying different approaches to greatly improve accuracy of the extracted information.
ALEX employs pattern-matching algorithms, mildly context-sensitive grammars adequate to model natural language structures, syntax analysis as well as pre-built dictionaries of related terms in order to parse the data.
A combination of approaches is also used by ALEX when identifying candidates’ skills in context. On the first pass ALEX extracts all probable skills from a resume. In addition ALEX provides an option of normalizing and optimizing all probable (“raw”) skills by conforming the skills to the terms in a compiled predefined dictionary of valid standard skills (terms). The dictionary is a vertical hierarchy where each hierarchy supports an industry such as IT, Medical, Finance, etc. Each hierarchy is a tree branch where each child is a “sub product” of its’ parent. For instance Oracle 9i is a child of Oracle and Oracle is a child Relational Database which it turn is a child under the top parent IT Industry. In this scenario when Oracle9i is mentioned on a resume Oracle, Relational Database and IT Industry are all added to the skills associated with that candidate. This option allows for extremely precise searches on candidate skills. In addition to vertical hierarchies the skills dictionaries normalize skills by defining synonyms for each main skill. For instance MS Word, MS-Word and Word are all synonyms of the main skill defined as Microsoft Word. Only the main skill will appear on the ALEX’s output.
In addition ALEX will attempt to expand on the number of extracted skills by searching for additional skills in the context of Job Descriptions (Employment Section of a resume) in resumes that generated less than 10 skills on the first pass. These additional skills are normalized and optimized unconditionally. These skills can be searched within client-defined industry branches of the Hierarchy tree. For instance a client may ask for only Medical terms to be searched on in this context as opposed to any terms.