/reader-embedding-api

Primary LanguageJupyter NotebookMIT LicenseMIT

reader-embedding-api

Toolforge interface to allow users to explore articles related to a given topic based on reader interest on Wikipedia articles.

UI based on the following template: https://github.com/wikimedia/research-api-interface-template

Endpoints

Possible arguments:

  • qid: wikidata item ot query; example:qid=Q81068910 (required)
  • n: number of related items to return (optional, default: 10, max: 100); example: n=10
  • lang: return article-title in corresponding wikipedia if it exists (optional, default: en); example: lang=en|de to get article-titles for enwiki and dewiki
  • threshold: threshold for similarity score, i.e. only return items that are above a certain similarity threshold (optional, default=0.0); similarity goes from 0.0 (not very similar) to 1.0 (identical); example: threshold=0.8
  • showurl: whether to show the url of the wikidata-item and the articles for easier exploration (optional, default=False); example: showurl=True
  • filter: filter items whose label contains the substrings (optional, default=None); example filter=covid|corona filters all wikidata items for which the lowercased English label contains the substrings 'covid' or 'corona'.

Example-query on the toolforge instance:

https://reader.toolforge.org/api/v1/reader/nn?qid=Q81068910

Additional notes

./code contains the scripts to generate the embedding. The actual trained model is hosted on cloud-vps due to memory-requirements. Thus, here we point to that endpoint.

License

The source code for this interface is released under the MIT license.

Screenshots of the results in the API may be used without attribution, but a link back to the application would be appreciated.