Afternoon session - NESCent

SKOS
-
2
-
HIVE

UNT workshop

Introductions

Craig Willis (craig.willis@unc.edu)

Afternoon Session Schedule


Overview


Using HIVE as a service


Installing and configuring HIVE


Using HIVE Core API


Understanding HIVE Internals


HIVE supporting technologies


Developing and customizing HIVE


Block 1: Introduction

Workshop Overview


Schedule


Interactive, less structure


Hands
-
on (work together)


Activities:


Installing and configuring HIVE


Programming examples (HIVE Core API, HIVE REST API)



Background and Interests


What are you most interested in getting out of this part of the
workshop?


What is your background?


Cataloging, indexing, and classification


Programming and databases


Systems administration


What is your level of familiarity with the following technologies?


Java, Tomcat, Lucene


REST


RDF, SPARQL, SKOS, Sesame



HIVE Technical Overview


HIVE consists of many technologies combined to provide a
framework for vocabulary services


System for management of multiple controlled vocabularies in
SKOS/RDF format


Java
-
based web services can run in any Java application server


Demonstration website (
http://hive.nescent.org/
)


Google Code project (
http://code.google.com/p/hive
-
mrc/
)

Architecture

HIVE Architecture


SPARQL:

RDF query language (W3C recommendation)


REST
:
Web
-
based API and software architecture


Triple store
:
Database for the storage and retrieval of RDF data.
Supports queries using SPARQL.


Sesame
: Open source triple store


Elmo
: Sesame API for common ontologies (OWL, Dublin Core,
SKOS)


Lucene
: Java
-
based search engine


KEA++
: Algorithm and Java API for automatic subject suggestions
from controlled vocabularies.


HIVE Functions


Conversion of vocabularies to SKOS


Rich internet application (RIA) for browsing and searching
multiple SKOS vocabularies


Java API and REST application interfaces for programmatic
access to multiple SKOS vocabularies


Support for natural language and SPARQL queries


Automatic keyphrase indexing using multiple SKOS
vocabularies. HIVE supports two indexers:


KEA++ indexer


Basic Lucene indexer

Block 2:

Using HIVE as a service

Using HIVE as a Service


HIVE web application


http://hive.nescent.org/


Developed by Jose Perez
-
Aguera, Lina Huang


Java servlet, Google Web Toolkit (GWT)


http://code.google.com/p/hive
-
mrc/wiki/AboutHiveWeb


HIVE REST service


http://hive.nescent.org/rs


Developed by Duane Costa, Long
-
Term Ecological Research Network


http://code.google.com/p/hive
-
mrc/wiki/AboutHiveRestService


Activity: Calling HIVE
-
RS


Demonstrate calling the HIVE
-
RS web service (Java)

Block 3:

Install and Configure HIVE

Installing and Configuring

HIVE


Requirements


Java 1.6


Tomcat
(HIVE is currently using 6.x)


Detailed installation instructions:


http://code.google.com/p/hive
-
mrc/wiki/InstallingHiveWeb


http://code.google.com/p/hive
-
mrc/wiki/InstallingHiveRestService



Installing and Configuring

HIVE
-
web


Detailed installation instructions (hive
-
web)


http://code.google.com/p/hive
-
mrc/wiki/InstallingHiveWeb


Quick start (hive
-
web)


Download and extract Tomcat 6.x


Download and extract latest hive
-
web war


Download and extract sample vocabulary


Configure hive.properties and agrovoc.properties


Start Tomcat


http://localhost:8080/


Properties files


hive.properties


Specifies enabled vocabularies and selected indexing algorithm


http://code.google.com/p/hive
-
mrc/source/browse/trunk/hive
-
web/war/WEB
-
INF/conf/hive.properties



<vocabulary>.properties


Specifies location of vocabulary databases/indexes on the local filesystem


http://code.google.com/p/hive
-
mrc/source/browse/trunk/hive
-
web/war/WEB
-
INF/conf/lcsh.properties


Installing and Configuring

HIVE
-
web from source


Detailed installation instructions (hive
-
web)


http://code.google.com/p/hive
-
mrc/wiki/DevelopingHIVE


http://code.google.com/p/hive
-
mrc/wiki/InstallingHiveWeb


Requirements


Eclipse IDE for J2EE Developers


Subclipse plugin


Google Eclipse Plugin


Apache Ant


Google Web Toolkit 1.7.1


Tomcat 6.x

Installing and Configuring

HIVE REST Service


Detailed installation instructions (hive
-
rs)


http://code.google.com/p/hive
-
mrc/wiki/InstallingHiveRestService


Quick start (hive
-
rs)


Download and extract latest webapp


Download and extract sample vocabulary


Configure hive.properties


Start Tomcat





Importing SKOS Vocabularies


http://code.google.com/p/hive
-
mrc/wiki/ImportingVocabularies


Note memory requirements for each vocabulary


http://code.google.com/p/hive
-
mrc/wiki/HIVEMemoryUsage


java
–
Xmx1024m
-
Djava.ext.dirs=path/to/hive/lib

edu.unc.ils.mrc.hive.admin.AdminVoc
abularies [/path/to/hive/conf/] [vocabulary] [train]

Block 4:

Using the HIVE Core Library

HIVE Core Interfaces

HIVE Core Packages

edu.unc.ils.mrc.hive.api

Main interfaces and implementations


edu.unc.ils.mrc.hive.converter


SKOS converters (MeSH, ITIS, NBII,
TGN)


edu.unc.ils.mrc.hive.lucene


Lucene index creation and searching


edu.unc.ils.mrc.hive.ir.tagging


KEA++ and “dummy” tagger
implementations


edu.unc.ils.hive.api


SKOSServer
:



Provides access to one or more vocabularies


SKOSSearcher
:



Supports searching across multiple vocabularies


SKOSTagger
:


Supports tagging/keyphrase extraction across multiple vocabularies


SKOSScheme
:


Represents an individual vocabulary (location of vocabulary on file
system)



SKOSServer


SKOSServer is the top
-
level class used to initialize the
vocabulary server.


Reads the
hive.properties
file and initializes the
SKOSScheme (vocabulary management), SKOSSearcher
(concept searching), SKOSTagger (indexing) instances based
on the vocabulary configurations.


edu.unc.ils.mrc.hive.api.SKOSServer


TreeMap<String, SKOSScheme> getSKOSSchemas();


SKOSSearcher getSKOSSearcher();


SKOSTagger getSKOSTagger();


String getOrigin(QName uri);

SKOSSearcher


Supports searching across one or more configured
vocabularies.


Keyword queries using Lucene, SPARQL queries using
OpenRDF/Sesame


edu.unc.ils.mrc.hive.api.SKOSSearcher


searchConceptByKeyword(uri, lp)


searchConceptByURI(uri, lp)


searchChildrenByURI(uri, lp)


SPARQLSelect()

SKOSTagger


Keyphrase extraction
using multiple vocabularies


Depends on setting in
hive.properties


edu.unc.ils.mrc.hive.api.SKOSTagger


“dummy” or “KEA”


List<SKOSConcept> getTags(String text, List<String>
vocabularies, SKOSSearcher searcher);

SKOSScheme


Represents an individual vocabulary, based on settings in
<vocabulary>.properties


Supports querying of statistics about each vocabulary
(number of concepts, number of relationships, etc).

Activity


Demonstrate a simple Java class that allows the user to query
for a given term.


Demonstrate a simple Java class that can read a text file and
call the tagger.

Block 5:

Understanding HIVE Internals

Architecture

Data Directory Layout


/usr/local/hive/hive
-
data


vocabulary/


vocabulary.rdf


SKOS RDF/XML


vocabularyAlphaIndex

Serialized map


vocabularyH2


H2 database (used by KEA)


vocabularyIndex


Lucene Index


vocabularyKEA


KEA model and training data


vocabularyStore


Sesame/OpenRDF store


topConceptIndex


Serialized map of top concepts

Keyword

Search

Indexing

HIVE Internals: Data Models


Lucene Index
: Index of SKOS vocabulary (view with Luke)


Sesame/OpenRDF Store
: Native/Sail RDF repository for
the vocabulary


KEA++ Model:
Serialized KEAFilter object


H2 Database:
Embedded DB contains SKOS vocabulary in
format used by KEA. (Can be queried using H2 command
line)


Alpha Index:
Serialized map of concepts


Top Concept Index:
Serialized map of top concepts

HIVE Internals: HIVE Web


GWT Entry Points:


HomePage


ConceptBrowser


Indexer


Servlets


VocabularyService: Singleton vocabulary server


FileUpload: Handles the file upload for indexing


ConceptBrowserServiceImpl


IndexerServiceImpl

HIVE Internals: HIVE
-
RS


Java API for RESTful Web Services (JAX
-
RS)


Classes


ConceptsResource:


SchemesResource



Block 6:

HIVE Supporting Technologies

HIVE supporting technologies


Lucene

http://lucene.apache.org


Sesame

http://www.openrdf.org/


KEA

http://www.nzdl.org/Kea/


H2


http://www.h2database.com/


GWT

http://code.google.com/webtoolkit/

Activity


Explore Lucene index with Luke


http://luke.googlecode.com/


Explore Sesame store with SPARQL


http://www.xml.com/pub/a/2005/11/16/introducing
-
sparql
-
querying
-
semantic
-
web
-
tutorial.html


http://www.cambridgesemantics.com/2008/09/sparql
-
by
-
example/

Block 7:

Customizing HIVE

Obtaining Vocabularies


Several vocabularies can be freely downloaded


Some vocabularies require licensing


HIVE Core includes converters for each of the supported
vocabularies.


List of HIVE vocabularies
http://code.google.com/p/hive
-
mrc/wiki/VocabularyConversion


Converting Vocabularies to SKOS


Additional information


http://code.google.com/p/hive
-
mrc/wiki/VocabularyConversion


Each vocabulary has different requirements


LCSH

Available in SKOS RDF/XML

NBII

Convert from XML to SKOS RDF/XML (SAX)

ITIS

Convert from RDB (MySQL) to SKOS RDF/XML

TGN

Convert from flat
-
file to SKOS RDF/XML

LTER

Available in SKOS RDF/XML

AGROVOC

Available in SKOS RDF/XML

MeSH

Convert from XML to SKOS RDF/XML (SAX)

Converting Vocabularies to SKOS


A Method to Convert Thesauri to SKOS (
van Assem
et al)


Prolog implementation


IPSV, GTAA, MeSH


http://thesauri.cs.vu.nl/eswc06/


Converting MeSH to SKOS for HIVE


Java SAX
-
based parser


http://code.google.com/p/hive
-
mrc/wiki/MeshToSKOS

LTER Sample Service

http://scoria.lternet.edu:8080/lter
-
hive
-
prototypes


Block 8:

KEA++

About KEA++


http://www.nzdl.org/Kea/


Algorithm and open
-
source Java library for extracting
keyphrases from documents using SKOS vocabularies.


Developed by Alyona Medelyan (KEA++), based on earlier
work by Ian Whitten (KEA) from the Digital Libraries and
Machine Learning Lab at the University of Waikato, New
Zealand.


Problem:
How can we automatically identify the topic of
documents?

Automatic Indexing


Free keyphrase indexing (KEA)


Significant terms in a document are determined based on intrinsic properties
(e.g., frequency and length).


Keyphrase indexing (KEA++)


Terms from a controlled vocabulary are assigned based on intrinsic
properties.


Controlled indexing/term assignment:


Documents are classified based on content that corresponds to a controlled
vocabulary.


e.g., Pouliquen, Steinberger, and Camelia (2003)

Medelyan, O. and Whitten I.A. (2008). “Domain independent automatic keyphrase indexing with small training sets.”
Journal of the American Society for Information Science and Technology
, (59) 7: 1026
-
1040).

KEA++ at a Glance


KEA++ uses a machine learning approach to keyphrase extraction


Two stages:


Candidate identification: Find terms that relate to the document’s
content


Keyphrase selection: Uses a model to identify the most significant terms.

KEA++: Candidate
identification


Parse tokens based on whitespace and punctuation


Create word n
-
grams based on longest term in CV


Stem to grammatical root (Porter)


Stem terms in vocabulary (Porter)


Replace non
-
descriptors with descriptors using CV relationships


Match stemmed n
-
grams to vocabulary


KEA++: Candidate
identification

Original

Stemmed

“information organization”

“inform organ”

“organizing information”

“inform organ”

“informative organizations”

“inform organ”

“informal organization”

“inform organ”

Stemming is not perfect ...

KEA++: Feature definition


Term Frequency/Inverse Document Frequency


Frequency of a phrase’s occurrence in a document with frequency in general
use.


Position of first occurrence:


Distance from the beginning of the document. Candidates with high/low
values are more likely to be valid (introduction/conclusion)


Phrase length:


Analysis suggests that indexers prefer to assign two
-
word descriptors


Node degree:


Number of relationships between the term in the CV.


DummyTagger


Primarily intended as baseline for analysis of KEA++


Uses LingPipe for part
-
of
-
speech identification (limits
indexing to certain parts of speech)


Uses Lucene vocabulary index


Simple TF*IDF implementation


Configurable in
hive.properties

Who’s Using HIVE


NESCENT/Dryad


Evaluating HIVE for automatic term suggestion from multiple
vocabularies for scientific article metadata.


Long
-
Term Ecological Research Network (LTERNet)


http://scoria.lternet.edu:8080/lter
-
hive
-
prototypes/


Prototype application for automatic term suggestion for EML
metadata files.


Library of Congress Web Archives


Evaluating HIVE for automatic term suggestion for web archive
(WARC) files




Plans


Automatic updates to vocabularies


Integration of other concept extraction algorithms


Maui


Dryad integration


Other


Maven integration


Spring integration


Data directory and property file restructuring


Concept browser updates



Discussion


Pros and Con


HIVE Core vs. HIVE Web vs. HIVE
-
RS


Brainstorm applications that could benefit from HIVE, discuss
implementations



Credits


Ryan Scherle


José Ramón Pérez Agüera


Lina Huang


Alyona Medelyan


Ian Whitten

Questions /Comments

Craig Willis


craig.willis@unc.edu