Description of the Corpus

The corpus used for Task A consists of a set of job titles in three languages: English, Spanish and German, from different job domains and professional sectors. These job titles have been collected and processed in order to facilitate the identification and comparison of equivalent titles across languages.

The training corpus has been generated using public terminologies, ensuring that the job titles are representative of a wide range of job domains and aligned with standard market terminology.

On the other hand, the validation and test corpora have been annotated by domain experts, following well-defined guidelines to ensure consistency and quality of labels. This annotation process, performed with specialized tools, included several stages of quality control to ensure that the labels were accurate and that the annotated titles accurately reflected the relationships between the different languages in a work environment.

Data:

1. Training Set: The training data is provided in a tabular format with three columns:

   • family_id: The ISCO family id representing the group to which the job identifier belongs.
   • id: An ESCO identifier indicating the origin of the pair’s job titles.
   • jobtitle_1: The first job title in the pair.
   • jobtitle_2: A second job title related to jobtitle_1.

   Each dataset is provided in separate files for each language involved in the task. The files are named according to the language, with the following format:

   • taskA_training_en.tsv: Contains related job titles in English.
   • taskA_training_es.tsv: Contains related job titles in Spanish.
   • taskA_training_de.tsv: Contains related job titles in German.

   An example of the content of these files is shown below:

2. Validation Set: The validation set is structured into three diferent files: queries, corpus elements and q_rels, and is provided separately for each language.

   • Queries: The queries file contains the following fields:

     • q_id: A unique identifier for the query.
     • jobtitle: The job title used as the query.

   • Corpus Elements: The corpus elements file contains the following fields:

     • c_id: A unique identifier for each corpus element.
     • jobtitle: The job title present in the corpus.

   • qrels: This file defines the relationship between the query and the corpus elements. It does not include a column header, but one is shown here for illustrative purposes.

     • q_id: The identifier of the query.
     • iter: A reserved field (always 0).
     • c_id: The identifier of the corresponding corpus element.
     • relevance: A binary score (0 or 1) indicating the relevance of the corpus element to the query, where 1 signifies relevant and 0 non-relevant.

   We will provide validation set in english, spanish, german and chinese.

   Example of the content of these files for english:

   queries

   q_id	jobtitle
   1	3d animator

   corpus_elements

   c_id	jobtitle
   1	animation artist
   2	3d character animator
   3	character technical director
   4	character designer
   5	animation lead
   6	3d generalist
   7	animator
   8	character rigger
   9	character animator

   q_rels

   q_id	iter	c_id	relevance
   1	0	2	1
   1	0	3	1
   1	0	4	1
   1	0	5	1
   1	0	6	1
   1	0	7	1
   1	0	8	1
   1	0	9	1

3. Test Set: The test set consists of two components, which are designed to evaluate system predictions based on language and job title retrieval tasks. The participant should generate a q_rels based on the queries and corpus elements provided.

   • Queries: Contains the following fields:

     • q_id: A unique identifier for the query.
     • jobtitle: The job title used as the query.
     • lang: The language of the corpus element’s job title.

   • Corpus Elements: Contains:

     • q_id: A unique identifier for each corpus element.
     • jobtitle: The job title from the corpus element.
     • lang: The language of the corpus element’s job title.

Summary:

The dataset is designed to support job title-based skill prediction tasks in English across various job domains and professional sectors. It includes job titles and associated skills collected and processed to facilitate the training of models to solve this task.

As with Task A, the training data uses public terminologies to represent a broad spectrum of job domains, while the validation and test sets are annotated by domain experts. This expert annotation follows strict guidelines and quality control measures to ensure consistent labeling and accurate representation of job-title-to-skill relationships.

Data:

1. Training Set:

For generating the training data for Task B, the information available in ESCO has been used. We have prepared the training data in three separate files: job2skill.tsv, jobid2terms.json and skillid2terms.json.

• job2skill.tsv: This file has been curated to include the most representative skills for each job title in ESCO. A filtering process has been applied to the number of skills per job title to avoid outliers. This file contains three columns:

  • job_id: ESCO identifier for the job position.
  • skill_id: ESCO identifier for the skill.
  • rel_type:Indicator specifying whether the skill_id is essential or optional for a specific job_id. It can have the value “essential” or “optional.”

  An example of the content of this file is shown below:

  job_id	skill_id	rel_type
  http://data.europa.eu/esco/occupation/f2b15a0e-e65a-438a-affb-29b9d50b77d1	http://data.europa.eu/esco/skill/8b94aa1e-89c9-459d-b3b4-1dfab8dec2df	essential
  http://data.europa.eu/esco/occupation/f2b15a0e-e65a-438a-affb-29b9d50b77d1	http://data.europa.eu/esco/skill/f84a433f-34f1-4083-b0a3-24802623509c	essential
  http://data.europa.eu/esco/occupation/f2b15a0e-e65a-438a-affb-29b9d50b77d1	http://data.europa.eu/esco/skill/fd33c66c-70c4-40e6-b87c-5495bd3bf26e	optional

• jobid2terms.json: This JSON file contains job_id identifiers used in the training set for Task A as keys, and a list of valid lexical variants for each identifier as values.

  {
      "http://data.europa.eu/esco/occupation/f2b15a0e-e65a-438a-affb-29b9d50b77d1": [
          "application developer", "application programmer", "applications engineer",
          "application software developer", "battery software developer",
          "developer of software", "programmer", "soft developer",
          "software developer", "software developers", "software engineer",
          "software specialist", "solutions developer"
      ]
      ...
  }
  

• skillid2terms.json: This JSON file contains skill_id identifiers as keys, and a list of valid lexical variants for each identifier as values.

  {
      "http://data.europa.eu/esco/skill/f84a433f-34f1-4083-b0a3-24802623509c": [
          "web services", "web services systems"
      ],
      "http://data.europa.eu/esco/skill/fd33c66c-70c4-40e6-b87c-5495bd3bf26e": [
          "design user interface"
      ]
  }
  

2. Validation Set: The validation set is divided into three diferent files: queries, corpus elements and q_rels:

   • Queries: Contains the following fields:

     • q_id: A unique identifier for the query.
     • jobtitle: The job title used as the query.

   • Corpus Elements: Contains:

     • c_id: A unique identifier for each corpus element.
     • esco_uri: The ESCO URIs associated to c_id.
     • skill_aliases: The list aliases of the ESCO skill

   • q_rels: This file maps the relationship between the query and the corpus elements:

     • q_id: The identifier of the query.
     • iter: A reserved field (always 0).
     • c_id: The identifier of the corresponding corpus element.
     • relevance: A binary score (0 or 1) indicating the relevance of the corpus element to the query, where 1 signifies relevant and 0 non-relevant.

   Example of the content of these files:

   queries

   q_id	jobtitle
   dev_qb_jt_1	corporate governance analyst

   corpus_elements

   c_id	esco_uri	skill_aliases
   dev_cb_sk_1	http://data.europa.eu/esco/skill/1c460d2d-90c6-4fc9-ad49-febb6e15605a	[‘pricing plans’, ‘price strategies’, ‘pricing tactics’, ‘pricing strategies’, ‘pricing strategy’]
   dev_cb_sk_2	http://data.europa.eu/esco/skill/301a6581-e983-4bb6-8b31-b3ee2cbc2392	[‘putting out fires’, …, ‘fires putting out’]
   dev_cb_sk_3	http://data.europa.eu/esco/skill/a4881e54-6055-4e61-855a-0a56ced7cfa3	[‘online assessment’, ‘analysis of web strategy’, ‘web presence assessment’, ‘web strategy assessment’]
   dev_cb_sk_4	http://data.europa.eu/esco/skill/efda73b4-5212-40a7-b2f8-d2f754ffdf2b	[‘keeping up with trends’, ‘keep pace with trends’, ‘follow trends’, …, ‘keep up with trends’]
   dev_cb_sk_5	http://data.europa.eu/esco/skill/22a173f5-868c-4d82-87e6-beed500ce070	[‘prepare tax returns form’, …, ‘make tax returns forms ready’, ‘preparing tax returns forms’]
   dev_cb_sk_6	http://data.europa.eu/esco/skill/97b890ff-acd7-46ad-8d3a-4186f4d42bbf	[’tuning procedures’, …, ’tuning skills’, ’tuning techniques’]
   dev_cb_sk_7	http://data.europa.eu/esco/skill/d5c20065-1d1f-446b-8143-9d1e180c512b	[‘iconography methods’, ‘iconography’]

   q_rels

   q_id	iter	c_id	relevance
   dev_qb_jt_1	0	dev_cb_sk_1034	1
   dev_qb_jt_1	0	dev_cb_sk_1087	1
   dev_qb_jt_1	0	dev_cb_sk_1088	1
   dev_qb_jt_1	0	dev_cb_sk_1099	1
   dev_qb_jt_1	0	dev_cb_sk_1104	1
   dev_qb_jt_1	0	dev_cb_sk_1107	1
   dev_qb_jt_1	0	dev_cb_sk_1110	1
   dev_qb_jt_1	0	dev_cb_sk_1112	1

3. Test Set: The test set consists of two files, queries and corpus elements. The participant should generate a q_rels file as prediction based on the queries and corpus elements provided.

   • Queries: Contains the following fields:

     • q_id: A unique identifier for the query.
     • jobtitle: The job title used as the query.

   • Corpus Elements: Contains:

     • q_id: A unique identifier for each corpus element.
     • skill: The skill associated with the corpus element.