Release Notes

v2 (March 13, 2017)

Version 2 of the NLU API is here, and we have lots of new things for you!

First, we added a new feature to our Entity Linking Service: it determines the entity salience score. Salient entities are more relevant to the text, and not just mentioned in passing. The score is in the range of [0,1], where higher means more salient. Use the salience score to find the most important and central entities in your text.

Second, our Knowledge Graph Service is now fully multilingual, supporting English, German, Spanish and Chinese, same as our Entity Linking Service. We've also added a couple of new endpoints: you can now query for a single entity, multiple entities in a get request. More importantly, you can search for entities by GPS coordinates (latitude, longitude, and distance), finding entities that are located in the specified circle.

Third, we are exposing an experimental endpoint /factextraction (English only for now). In addition to the functionality provided by the entitylinking endpoint, it extracts a list of open facts found in the text in the format subject, relation, object. No longer does our API only find entities in your texts, but also the relations between them!

Schema changes:

Entity Linking Service

  • We introduced a new list of entities, which contains all unique entities that appear in the document. An entity consists of the following fields:
    • id - the id of the entity.
    • name - the human readable name of the entity in the requested language.
    • url - the Wikipedia url of the entity.
    • type - the most high-level type of the entity.
    • salience - the salience score of the entity signifying the importance of the entity in the text.
  • The entity object in the list of matches now contains only id and confidence. We moved the name and the url fields to the new list of entities described above.

Knowledge Graph Service

  • The entities list is now a (id, Entity) map, for faster lookup of entities in the result.
  • We've added the field type, giving the most high-level entity type.
  • We've removed the field name and replaced it with names, which is a (language, label) map, where language is one of the supported languages, and a label is an object which again consists the language code of the name and a value in the specified language.
  • We've removed the field description, and replaced it with descriptions, which is a (language, label) map, of the language codes and short descriptions in the supported languages of the entity.
  • We've added a field longDescriptions, which is a (language, label) map, of the language codes and long descriptions in the supported languages of the entity.
  • We've made the links field multilingual as well, following the pattern explained in the previous fields.
  • We've removed the field imageUrl, and replaced it with an object called image, which contains the url of the image, plus a list of licenses, which contain the name of the license and the url of the license, and an author which is represented as name and url.
  • The /categories endpoint also returns a (id, Category) map, for faster lookup of the categories in the result.
  • The category object is extended by a multilingual map of descriptions for the category.

v1 (December 19, 2016)

Our first stable release! Scalable and with a guaranteed uptime of 99%, including our new Enterprise Plan. We significantly improved the entity linking quality for German. More names are recognized and linked with higher accuracy.

Schema changes:

  • We integrated Wikidata and now use Wikidata entity identifiers for entity IDs. This allows you to easily enter the world of Linked Data, and also makes sure that entity IDs remain stable over time.
  • You can now add your own mentions as input using the annotatedMention field.

v1beta3 (September 20, 2016)

Our "Pro" plan is now available - you can now process more than 1,000 documents per month. Even better, in the beta period you will get a a 50% discount!

That's almost it, only one minor change that makes the Entity Linking and the Knowledge Graph Service more consistent:

  • Entities and categories are always referred to by id instead of kgId.

v1beta2 (July 26, 2016)

Hello! ¡Hola! 你好! Hallo! - our API now understands Spanish, Chinese, and German. In addition to English of course.

The new Beta version v1beta2 contains significant additional changes to the API under the hood. The most important things we’ve done:

  • We’ve changed the way the API is exposed to the public: We prepared the API for offering production-ready usage plans very soon, and offer a full-featured developer portal with documentation, analytics, and developer self-services such as OAuth 2 client credentials management.
  • We’ve unified the API endpoints. Our services are now accessible from https://api.ambiverse.com/v1beta2/<service>.
  • The Entity Linking Service now also serves image URLs in the responses so that you can visually display entities.
  • We're supporting multilingual analysis (EN, DE, ES, ZH) and automatic language identification of the input text if no language is specified.

v1beta1 (May 18, 2016)

The initial release of the Ambiverse Natural Language Understanding API. It combines two essential tasks. Given a text, our Entity Linking Service recognizes names and matches them onto canonical entities (e.g., individuals, organizations or places) registered in the Wikipedia-derived YAGO knowledge base. You can then retrieve additional information about these entities using the Knowledge Graph Service, uncovering the semantic meaning of the text.