使用语言模型从在线职位空缺中提取就业地区.pdf

1、Using language models for extracting regions of employment from online job vacanciesGdask,Poland I 04 Feb 2025Adam TsakalidisCedefop Expert(Skills Intelligence&Foresight)Web Intelligence Network Conference From Web to DataAdam Tsakalidis&Antonio RanieriIntroduction2Task:Extracting regions of employm

2、ent from Online Job Advertisements(OJAs)IntroductionTask:Extracting regions of employment from Online Job Advertisements(OJAs)Why this mattersOJAs:real-time monitoring of the labour marketRegions:fine-grained resolution3IntroductionTask:Extracting regions of employment from Online Job Advertisements

3、(OJAs)Why this mattersOJAs:real-time monitoring of the labour marketRegions:fine-grained resolutionData:Skills-OVATEESTAT+CedefopHundreds of millions of OJAsMultiple languagesEU27+coverage2018-24Classifications:occupations(ISCO),skills(ESCO),4IntroductionTask:Extracting regions of employment from On

4、line Job Advertisements(OJAs)Why this mattersOJAs:real-time monitoring of the labour marketRegions:fine-grained resolutionData:Skills-OVATEESTAT+CedefopHundreds of millions of OJAsMultiple languagesEU27+coverage2018-24Classifications:occupations(ISCO),skills(ESCO),In this work:-Greek language-NUTS-2

5、 region56ChallengesPlan A:Using commercial LLMs-Data sensitivity-CostChallengesPlan A:Using commercial LLMsPlan B:Prompting in-house LLMs-Cost-Scalability78ChallengesPlan A:Using commercial LLMsPlan B:Prompting in-house LLMsPlan C:Typical ML/NLP process(a)Manual annotation of two datasets(train/test

6、)(b)Training a model on the training data(c)Evaluate model on the test data9ChallengesPlan A:Using commercial LLMsPlan B:Prompting in-house LLMsPlan C:Typical ML/NLP process(a)Manual annotation of two datasets(train/test)(b)Training a model on the training data(c)Evaluate model on the test data-Time