X-Link

Overview

X-Link is a fully configurable (Linke Data-based) Named Entity Extraction (NEE) tool which allows the user/developer to easily define the categories of entities that are interesting for the application at hand by exploiting one or more (online) Semantic Knowledge Bases (Linked Data). The user is also able to update a category and specify how to semantically link and enrich the identified entities. This enhanced configurability allows X-Link to be lightly configured for different contexts, for building domain-specific applications (e.g. for identifying drugs in a medical search system, for annotating and exploring fish species in a marine-related web page, etc.).

X-Link is based on Gate ANNIE and supports both gazetteers (lists of names) and natural language processing functions. Gate ANNIE is a ready-made information extraction system which contains several components (e.g. Tokeniser, Gazetteer, Sentence Splitter, Orthographic Coreference, etc.). We have extended Gate ANNIE in order to be able to create a new supported category and update an existing one (using gazetteers) by exploiting the Linked Data.

Currently, X-Link exports the results in XML and CSV.

Key features

Currently X-Link supports the analysis of plain text files, HTML pages, Microsoft Word and Powerpoint files (.doc, .docx, .ppt and .pptx), PDF files, and XML-based files (e.g. XML and RDF files). At first it reads the contents of the corresponding document and performs a "cleaning" task, i.e. it removes useless text (e.g. HTML tags in a Web page or Meta elements in a Microsoft Word file). Then, it applies NEE in the cleaned contents of the document.

X-Link starts by reading an initial configuration which is stored in a properties file. It also implements functions that allow the user/developer to configure the system, e.g. through an administrator API. Specifically, the following functions are currently supported:

• Add a new category, using one or more lists of entities, one or more instances resource classes or one or more instances SPARQL queries.
• Update an existing category, using one or more lists of entities, one or more instances resource classes or one or more instances SPARQL queries. The user can either totally replace a category (i.e. remove the old entities and add the new ones) or just add the new entities.
• Remove an existing category.
• Change the dispayed name of an existing category (i.e rename).
• Set/change the underlying Knowledge Bases
• Set/change how to query the underlying Knowledge Bases for linking the identified entities.
• Set/change how to query the underlying Knowledge Bases for enriching the identified entities.
• Set/change the active categories, i.e. the categories for which X-Link identified entities.

Functionality

Load the entity mining component, get the currently supported categories of entities and print their names

EntityMiningComponent emc = new GateEntityMiningComponent("C:/XLinkGateComponent"); 
emc.startup(); 
 
ArrayList<Category> availableCategories = emc.getAvailableCategories();
 
System.out.println("# Available categories: ");
for (Category category : availableCategories) {
   System.out.println(category.getName());
}

Create an instance of X-Link, set the accepted (active) categories of entities and perform entity mining in a Web page

XLink xlink = new XLink();
xlink.setEntityMiningComponent(emc);
HashSet<String> acceptedCategoryNames = new HashSet<String>(); 
acceptedCategoryNames.add("species");
TextExtractor extractor = new WebPageTextExtractor("http://en.wikipedia.org/wiki/Yellowfin_tuna");
xlink.retrieveEntities(extractor, acceptedCategoryNames);

Link the identified entities with semantic resources (i.e. URIs)

xlink.matchEntities();

Enrich the entity URIs with semantic information (i.e. RDF triples)

• Enrich all entity URIs with their outgoing properties:

xlink.enrich(PropertiesType.OUTGOING);

• Enrich the entity URIs of the category 'Species' with their outgoing properties:

xlink.enrich(emc.getCategory("species"), PropertiesType.OUTGOING);

Infer the connectivity of the identified entities

(to be completed)

Stop the entity mining component and print the identified entities, together with the matching URIs

emc.shutdown();

ArrayList<Entity> entities = xlink.getEntities(); // Gets the detected entities (together with all their information).
 
System.out.println("# Detected entities: ");
for (Entity entity : entities) { // Print the mane characteristics of the detected entities.
   System.out.println("Entity name: " + entity.getName());
   System.out.println("Category: " + entity.getCategoryName());
   System.out.println("Matching URIs: " + entity.getMatchingURIs());
   System.out.println("-----");
}

Export the results

ResultExporter exp1 = new XMLExporter("C:/x-link/results/results.xml", entities);
exp1.exportResults();
       
ResultExporter exp2 = new TXTExporter("C:/x-link/results/results.txt", entities);
exp2.exportResults();

ResultExporter exp3 = new CSVExporter("C:/x-link/results/results.csv", entities);
exp3.exportResults();

Configurability

The 'GATE Annie' Entity Mining Component

(to be completed)

The 'x-link.properties' file

(to be completed)

Configuring (programmatically) X-Link

X-Link starts by reading an initial configuration which is stored in a properties file but also offers the appropriate functions that facilitate its configuration (e.g. through an API) in a preprocessing step or even while a corresponding service is running.

Adding a new category of entities

Giving a list of entity names

Category categoryToAdd = new Category("North Aegean Greek Islands");

TreeSet<String> names = new TreeSet<String>();
names.add("Lesvos");
names.add("Chios");
names.add("Samos");
names.add("Limnos");
names.add("Ikaria");
names.add("Samothraki");
names.add("Agios Eustratios");
names.add("Psara");
names.add("Fournoi");
names.add("Oinouses");
names.add("Thymaina");
names.add("Antipsara");
names.add("Pasas");
names.add("Agios Minas");
names.add("Samiopoula");
 
categoryToAdd.setNamedEntities(names);   
emc.addNewCategory(categoryToAdd);      

Giving a resource class

Category categoryToAdd = new Category("Fish Species");
categoryToAdd.setResourceClass("http://dbpedia.org/ontology/Fish");
categoryToAdd.setEndpoint("http://dbpedia.org/sparql");
categoryToAdd.retrieveNamedEntitiesByResourceClass();

emc.addNewCategory(categoryToAdd);

Giving a SPARQL query

Category categoryToAdd = new Category("Fish Species");
categoryToAdd.setEndpoint("http://dbpedia.org/sparql");
categoryToAdd.setSparqlQuery("select distinct str(?name) where { ?uri a <http://dbpedia.org/ontology/Fish> . ?uri rdfs:label ?name }");
categoryToAdd.retrieveNamedEntitiesByQuery();
emc.addNewCategory(categoryToAdd);

or (in case the SPARQL query exists in a file)

Category categoryToAdd = new Category("Fish Species");
categoryToAdd.setEndpoint("http://dbpedia.org/sparql");
categoryToAdd.setSparqlQueryFilepathOfEntities("C:/sparql/getFishSpeciesQuery.sparql");
categoryToAdd.retrieveNamedEntitiesByQuery();
emc.addNewCategory(categoryToAdd);

Updating the entities of a category

Giving a list of entity names

Set<String> newEntities = new TreeSet<String>();
newEntities.add("Patmos");
newEntities.add("Crete");
newEntities.add("Karpathos");
     
emc.updateCategoryBySet("Greek Islands", newEntities);

Giving a resource class

 String endpoint = "http://lod.openlinksw.com/sparql";
 String resourceClass = "http://umbel.org/umbel/rc/Fish";

 emc.updateCategoryByResourceClass("Fish Species", endpoint, resourceClass);

Giving a SPARQL query

String endpoint = "http://dbpedia.org/sparql";
String sparqlQuery = "select distinct str(?name) where { ?uri a <http://dbpedia.org/ontology/Fish> . ?uri rdfs:label ?name FILTER(lang(?name)='es') }"
emc.updateCategoryByQuery("water areas" , endpoint, sparqlQuery);

or (in case the SPARQL query exists in a file)

String endpoint = "http://dbpedia.org/sparql";
emc.updateCategoryByQuery("water areas" , endpoint, "C:/sparql/getWaterAreas.sparql");

Replacing the entities of a category

Giving a list of entity names

Set<String> newEntities = new TreeSet<String>();
newEntities.add("Patmos");
newEntities.add("Crete");
newEntities.add("Karpathos");
     
emc.replaceCategoryBySet("Greek Islands", newEntities);

Giving a resource class

 String endpoint = "http://lod.openlinksw.com/sparql";
 String resourceClass = "http://umbel.org/umbel/rc/Fish";

 emc.replaceCategoryByResourceClass("Fish Species", endpoint, resourceClass);

Giving a SPARQL query

String endpoint = "http://dbpedia.org/sparql";
String sparqlQuery = "select distinct str(?name) where { ?uri a <http://dbpedia.org/ontology/Fish> . ?uri rdfs:label ?name FILTER(lang(?name)='es') }"
 
emc.replaceCategoryByQuery("water areas" , endpoint, sparqlQuery);

or (in case the SPARQL query exists in a file)

String endpoint = "http://dbpedia.org/sparql";
emc.replaceCategoryByQuery("water areas" , endpoint, "C:/sparql/getWaterAreas.sparql");

Renaming a category

String newName = "Fish";
emc.renameCategory("Fish Species", newName);

Removing a category

emc.removeCategory("Fish Species");

Specifying how to link the identified entities with semantic resources (i.e. URIs)

SPARQL linking template query

For each supported category of entities, X-Link retain a SPARQL linking template query. The SPARQL linking template query describes the way X-Link must query the corresponding SPARQL endpoint for retrieving resources (i.e. URIs) that match a particular name of entity. Specifically, a SPARQL linking template query contains the character sequence <ENTITY> (including the < and >). At request time, the system reads the endpoint and the corresponding template query of the category in which the identified entity belongs, replaces each occurrence of <ENTITY> in the template query with the entity's name, and finally runs the query. For example, the following template query tries to find a resource of type (rdf:type) Fish whose label (rdfs:label) contains the name of the identified entity (ignoring case):

SELECT DISTINCT ?uri WHERE {
  ?uri rdf:type <http://dbpedia.org/ontology/Fish> .
  ?uri rdfs:label ?label FILTER(regex(str(?label), '<ENTITY>', 'i')) 
}

(to be completed)

Specifying how to enrich the identified entities with semantic information (i.e. RDF triples)

(to be completed)

Inspecting the connectivity of the identfied entities

(to be completed)