Pilveteenus Eesti struktrueerimata avalike andmete kuluefektiivseks töötlemiseks, linkimiseks ja päringute tegemiseks
Date
2015
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Selle projekti eesmärk on luua pilveteenus, mis võimaldaks struktueerimata avalike andmete töötlemist, selleks, et luua semantiline andmete (veebis olevatest dokumentidest leitud organisatsioonide, kohanimede ja isikunimede) ressursikirjeldusraamistiku - Resource Description Framework (RDF) - graaf, mis on ka masinloetav. Pilveteenus saab sisendiks veebiroomaja toodetud logifaili üle 3 miljoni reaga. Igal real on veebiaadress avalikule dokumendile, mis avatakse, loetakse ning kasutades - tööriista eestikeelsest tekstist nimeolemite leidmiseks- Estnltk-d, eraldatakse organisatsiooonide ja kohtade nimetused ja inimeste nimed. Seejärel lisatakse leitud nimed/nimetused RDF graafi, kasutades olemasolevat Pythoni teeki RDFlib. RDF graafis nimed/nimetused lingitakse nende veebiaadressidega, kus asub seda nime/nimetust sisaldav avalik dokument. Dokumendid arhiveeritakse lugemise hetkel neis olnud sisuga. Lisaks sisaldab teenus igakuist andmete ülekontrollimist, et tuvastada dokumentide muutusi ja vajadusel värskendada RDF graafe. Genereeritud RDF graafe kasutatakse SPARQL päringute tegemiseks, mida saavad teha kasutajad graafilise kasutajaliidese kaudu või masinad veebiteenust kasutades. Projekti oluline väljakutse on luua arhitektuur, mis töötleks andmeid võimalikult kiiresti, sest sisendfail on suur (test-logifailis on üle 3 miljoni rea, kus igal real olev URL võib viidata mahukale dokumendile). Selleks jooksutab teenus seal kus võimalik, protsesse paralleelselt, kasutades Google’i virtuaalmasinaid (Google Compute Engine) ja iga virtuaalmasina kõiki protsessoreid.
The aim of this project is to develop a cloud platform service for transforming Open Government Data to Linked Open Government Data. This service receives log file, created by web crawler, with URLs (over 3000000) to some open document as an input. It then opens the document, reads its content and with using "Open source tools for Estonian natural language processing" (Estnltk), finds names of locations, organizations and people. Using Psython library "RDFlib", these names are added to the Resource Description Framework (RDF) graph, so that the names become linked to the URLs that refer to the documents. In order to archive current state of accessed document, this service downloads all processed documents. The service also enables monthly updates system of the already processed documents in order to generate new RDF relations if some of the documents have changed. Generated RDFs are publicly available and the service includes SPARQL endpoint for userss (graphical user interface) and machines (web services) for cost-effective querying of linked entities from the RDF files. An important challenge of this service is to speed up its performance, because the documents behind these 3+ billion URLs may be large. To achieve that, parallel processes are run where possible: using several virtual machines and all CPUs in a virtual machine. This is tested in Google Compute Engine
The aim of this project is to develop a cloud platform service for transforming Open Government Data to Linked Open Government Data. This service receives log file, created by web crawler, with URLs (over 3000000) to some open document as an input. It then opens the document, reads its content and with using "Open source tools for Estonian natural language processing" (Estnltk), finds names of locations, organizations and people. Using Psython library "RDFlib", these names are added to the Resource Description Framework (RDF) graph, so that the names become linked to the URLs that refer to the documents. In order to archive current state of accessed document, this service downloads all processed documents. The service also enables monthly updates system of the already processed documents in order to generate new RDF relations if some of the documents have changed. Generated RDFs are publicly available and the service includes SPARQL endpoint for userss (graphical user interface) and machines (web services) for cost-effective querying of linked entities from the RDF files. An important challenge of this service is to speed up its performance, because the documents behind these 3+ billion URLs may be large. To achieve that, parallel processes are run where possible: using several virtual machines and all CPUs in a virtual machine. This is tested in Google Compute Engine