Lucene Web Service Parameters
All parameters accepted by DBpedia Spotlight are describing in the WADL (Web Application Description Language) file.
Following there are a description of each parameter used by the web service. Many of them are used in the filter and are described in that section.
text or url (Required): text or url to be annotated.
It is possible to remove unwanted annotations through the endpoints using filters parameters. These filters are available in /candidates and /annotate (applied after disambiguation step) endpoints.
It is a heuristic that seeks coreference in all text and infer the surface form. When is true, no other filter will be applied.
Available in: /candidates, /annotate
Parameter name: coreferenceResolution
Parameter type: boolean
Default value: true
Selects all entities that have a percentageOfSecondRank greater than the square of value informed.
Available in: /candidates, /annotate
Parameter name: confidence
Parameter type: number(double)
Default value: 0.1
Selects all entities that have a support greater than informed.
Available in: /candidates, /annotate
Parameter name: support
Parameter type: number(integer)
Default value: 10
Combined with policy parameter, select all entities that have the same type - if policy is whitelist. Otherwise - if policy is blacklist - select all entities that have not the same type.
Usage:
types=DBpedia:PopulatedPlaces,DBpedia:Thing
Available in: /candidates, /annotate
Parameter name: types
Parameter type: string
Combined with policy parameter, select all entities that match with the query result - if policy is whitelist. Otherwise - if policy is blacklist - select all entities that no match with the query result.
Available in: /candidates, /annotate
Parameter name: sparql
Parameter type: string
Spotters are algorithms that select all candidates for possible annotations. There are two kind of implementations. In the language-independent implementation, the candidates are generated by traversing a finite state automaton encoding all possible sequences of tokens that form known spot candidates.
In the language-dependent implementation, candidates are generated using three methods: 1. identifying all sequences of capitalized tokens, 2. identifying all noun phrases, prepositional phrases and multi word units, 3. identifying all named entities. Methods 2 and 3 are performed using Apache OpenNLP6 models for phrase chunking and Named Entity Recognition.
Available in: /candidates, /annotate, /spot
Parameter name: sparql
Parameter type: string
Default value: Default
Possible values:LingPipeSpotter, WikiMarkupSpotter, AtLeastOneNounSelector, CoOccurrenceBasedSelector, NESpotter, OpenNLPNGramSpotter, OpenNLPChunkerSpotter,KeaSpotter
You can change the strategy using for spotting by changing the value of the &spotter= parameter passed to our web service. Supported spotters:
- Default, picks the first spotter informed in the configuration file
- LingPipeSpotter, uses a dictionary of known names to spot
- AtLeastOneNounSelector, uses the dictionary and removes spots that do not contain a noun
- CoOccurrenceBasedSelector, uses the dictionary and removes spots that "look like" a non-entity (as trained by feature co-occurrence statistics)
- NESpotter, uses the OpenNLP default models for Named Entity Recognition (NER)
- KeyphraseSpotter, uses the Kea default models for Keyphrase Extraction
- WikiMarkupSpotter, assumes that another tool has performed spotting and encoded the spots as WikiMarkup
- SpotXmlParser, assumes that another tool has performed spotting and encoded the spots as SpotXml.
Available in: /candidates, /annotate, /spot
Parameter name: sparql
Parameter type: string
Default value: Default
Possible values: Document, Occurrences, CuttingEdge, Default
DBpedia Spotlight - Shedding Light on the Web of Documents
Project
Model backend
Developers
Google Summer of Code - GSoC