Detecting Uninformative Research Job Titles via Classifier Failures—Zero Shot Approach

Abstract

The aim of this paper is to introduce a novel approach to detecting "uninformative" job titles in research domain, i.e., detecting titles that convey little or no information about the focus and/or content of a particular job -like "Academic staff member AP/2", "PhD student position" etc. Such job titles decrease the success rate of job advertisements. The proposed approach belongs to zero shot approaches -it exploits only existing, easy accessible classification of jobs to research fields and it does not require any additional (manual) annotations. This work introduces an experimental corpus and provides preliminary results of our approach.