Our search engine performs combined search over structured data (triple stores) and full-text
Our current prototype uses the English Wikipedia as the full-text and Freebase as its ontology.
Here you can find a list with the most important features of Broccoli.
Features
| Interactive Suggestions | Broccoli provides interactive query suggestion at any time during query construction. The query for tall buildings, for example, suggestest other classes of the current result, particulary entity instances that satisfy the query and relations frequent among them. A query for buildings will suggest a relation called "Floors", a query for persons won't. | See in action |
| NLP Debug Info | Broccoli uses its own processing of natural language in order to decide which parts of a sentence semantically belong together. Click on the circle next to a document name in the large box with the hits to open a window with debug information. It contains various NLP features, including a full syntactic parse for the current excerpt. In the example link, click the circle next to "Edible plant stem" in the hit for Broccoli. | See in action |
| Order By Relation | Results are ranked by how well they match the query. In addition they can be reranked by the result of one relation attached to the root. Click on the small up and down arrow on the left of the little green relation boxes of the current query. This feature can, for example, be used to order buildings by their height. See the example link. | See in action |
| Keyword Queries | Broccoli can automatically interpret some natural language queries. This features is experimental and in no way a full system for question answering. | See in action |
| Standard Full-Text Search | Standard Full-Text Search is included as a special case. Documents are entities, too. Instead of the occurs-with relation that is used for typical entity centic queries, a relation has-occurrence of is used in queries for documents | See in action |