Distributional Semantics Over Semi-structured Data

点击翻译按钮获取中文摘要

Many applications involve both structured and unstructured data. For example, in a system that matches jobs to job seekers, the structured component includes the formal qualifications required, the location of the job, the job title, etc., whereas the unstructured component is the raw text description of the job itself and the textual parts of the candidate's resume. In this project, we will investigate how to incorporate both sources of information in order to build a semantic representation that can support useful inferences about the domain such as ranking candidates to interview for a position. Using a data set from Workhoppers, a company that matches job seekers to contract positions, we will develop statistical models of meaning that consider the words that are present in the input data as well as their position within the structure of the data in order to build a better profile of an entity such as a job opening or a job seeker. These meaning representations will then be applied to ranking job candidates, and to present a summary of a job candidate that highlights how their qualifications meet the job description. This research will benefit Workhoppers' matching algorithm by enabling more job candidates to be matched to an appropriate job. It could also benefit Canadian economy by accelerating the job search process.