Neighbors concept algorithm

Project

Implement the algorithms described in Answers Partitioning and Lazy Joins for Efficient QueryRelaxation and Application to Similarity Search by Sébastien Ferré in Jena A Java library for Semantic Web

Objectives

• Finish implementing the two algorithms
• Compare the results obtained by the early implementation made using OCaml
• Make all the necessary accommodations for the patch to be re-usable by other Jena users
• Create a Demo version, if possible in the form of a Graphical User Interface

Resources

• https://hal.archives-ouvertes.fr/hal-01945454/document : Publication describing the algorithms
• https://jena.apache.org/documentation/javadoc/jena/ : Jena Documentation for Base RDF Graphs
• https://jena.apache.org/documentation/javadoc/arq/ : Jena Documentation for ARQ, the SPARQL Engine
• https://openjfx.io/javadoc/12/ : JavaFX Documentation

User Documentation

Graphical User Interface

• Execute implementation.gui.NeighborsInterface.main() (done by the run_interface.sh script)
• Find your RDF File
  • Click on the Browse button and find your RDF file in the file explorer
  • or directly type the absolute path of the file on your system in the text field next to the button
• Select the right format for your file
• Click on Load RDF File
• In the choice box that appears on the right, select the uri of the node you want to find the neighbors of
  • If there are too many possible nodes, you can narrow it down by typing part of the uri you want in the text field next to the Filter Button

Redistributable

Jena and JavaFX are not included in this repository.
To use the launcher scripts, change variables in config to set up the path to your JDK, Jena and JavaFX
This project was developed using JDK 12.0.1, Jena 3.12.0 and JavaFX 12.0.1, make sure to use versions that are compatible with these

Four datasets are included in the repository :

• royal.ttl is a small sample dataset that describes basic familial links in the british royal family
• mondial.ttl, mondial.n3 and mondial-europe.n3 describe many geographical items (downloaded from https://www.dbis.informatik.uni-goettingen.de/Mondial/)

Downloads section on bitbucket (https://bitbucket.org/sebferre/conceptsofneighbours/downloads/) contains two "bundles" if you're only interested in specific uses and don't want to clone the whole repository

• GUI_only : the compiled code and launcher script, you will still need to configure your paths in the config file
• lib_only : compiled code in an archive to use as module in java

Main Class

• Every part of the main is explained in comments in the code
• The lines you might want to change are :
  • The one setting up the String filename which is the absolute path of the file containing the RDF data
  • The one setting up the String format change the format to the format of your file (see Jena Documentation for supported formats)
  • The one setting up the String uriTarget which is the uri of the node you want to find the neighbors of
  • The few ones at the beginning, defining the log4j logging levels

Current Version

Javadoc at https://nk-fouque.github.io/Neighbors-Concept/allclasses-index.html Run run_interface.sh to open interface

Content

• Class Main gives an example use of the partition algorithm

• Class NeighborsImplementation for some static function useful to Main

package algorithms

• Classes Partition and Cluster for the first Algorithm described in the publication
• Classes LazyJoin, MatchTreeRoot and MatchTreeNode for the second Algorithm described by the publication

package utils

• Class CollectionsModel using Java collections to store informations about the Model, notably in Hashmaps, very useful for their computational speed to avoid using ARQ
• Classes ListUtils, ElementUtils, TableUtils containing useful static functions respectively for java.util.List, org.apache.jena.sparql.syntax.Element and org.apache.jena.sparql.algebra.Table
• Classes Stopwatch and SingleStopwatchCollection a very simple class of stopwatches to compensate not being able to run Async Profiler on my computer
• Other misc. classes, see Javadoc

package gui

MVC representation for the Interface Demo
See classes Javadoc for details

Anytime Implementation

Manual Interruption

Both implementations use a version that cuts the algorithm in case of OutOfMemoryError or under some action of the user

• The Main class intercepts the SIGINT signal, naturally sent by IntelliJ when you click on the stop button (the second time it sends SIGKILL)
• The Interface has a Button to stop the algorithm

When the algorithm is interrupted, it finishes the current run of iterate() and moves every clusters to neighbors. It considers the algorithm over, printing/displaying everything it was supposed to.

• In the Main class, the process then stops
• In the interface, the process can be launched for another node

TimeOut

The Main class uses a back thread that cuts the algorithm after a certain amount of time

The Interface has a field to select a time limit for the algorithm (also done in a back thread), this limit cannot be set after starting the algorithm, the user will have to use the interrupt button

Results

• Works perfectly with any item of royal
• Works with mondial at depth 1 or with timeout