For the task of job recommendations, common practice is to recommend job postings to job seekers, and recently different embedding techniques have been applied to solve this task. A common way to represent job seekers and job postings as embeddings is to use the whole textual data of the job postings and job seekers' resumes. Instead of using the whole textual data, textual data source like job title, knowledge, skills, abilities can be used as well. In this work, we present findings of a preliminary offline study, where we explore the impact of utilizing different types of embeddings to recommend job seekers to given job postings, unlike the common practice of recommending job postings to job seekers. We explore the effectiveness of using job title based embeddings compared with the embeddings based on resume and job posting full-text descriptions. Using a dataset from JobIndex -Scandinavia's largest job portals and recruitment agencies- our experimental results show that representing job seekers and job postings as embeddings by using job title text only can be at least as informative as using the full-text descriptions for most of the cases.
|Tidsskrift||CEUR Workshop Proceedings|
|Status||Udgivet - 2021|
|Begivenhed||2021 Workshop on Recommender Systems for Human Resources, RECSYS IN HR 2021 - Amsterdam, Holland|
Varighed: 27 sep. 2021 → 1 okt. 2021
|Konference||2021 Workshop on Recommender Systems for Human Resources, RECSYS IN HR 2021|
|Periode||27/09/2021 → 01/10/2021|