Bachelor's thesis

Information retrieval for a web search engine using neural networks and classic methods

Key facts

Author
Lucas Imhof
University
Hochschule Merseburg
Faculty
Fachbereich Ingenieur- und Naturwissenschaften
First examiner
Prof. Dr. rer. pol. Uwe Schröter
Second examiner
Prof. Dr. rer. nat. habil. Dr. phil. Michael Schenke
Submitted
11 March 2019
Published
28 March 2019
Length
102 pages
Language
Deutsch
Publisher
Hochschulbibliothek, Hochschule Merseburg
PPN
1662878591
Licence
CC BY 4.0
Keywords
Information Retrieval · Retrieval-System · Suchmaschine · Bildersuche

Abstract

This bachelor's thesis deals with information retrieval for a web search engine. The quality of the data collected by information retrieval largely determines the efficiency of the search engine that is supposed to deliver results to a user from that data. Quality means that a user receives a result that is useful for the task at hand – data that matches the query both thematically and in content, and that is prioritised correctly depending on several factors discussed in the thesis. Image search is a particular challenge: it cannot simply match keywords, the content of the image has to be analysed and categorised by the retrieval system first.

In parallel to the written thesis, a prototype of a search engine was built. It consists of a crawler network that extracts information from web pages, images and documents and stores it in a database via a server, plus a search engine that lets you search for text fragments and images found by the crawler network. (Translation; the thesis itself is written in German.)

In numbers

102

pages

6

chapters

26

figures

3

tables

27

code listings

7

formulas

88

references

Table of contents

Chapter titles in the original German, page numbers as in the PDF.

  1. 1EinleitungS. 4
  2. 2GrundlagenS. 5
  3. 3Information RetrievalS. 10
  4. 4Künstliches neuronales NetzS. 27
  5. 5PrototypS. 49
  6. 6SchlusswortS. 89
  7. Verzeichnisse (Abkürzungen, Glossar, Formeln, Listings, Abbildungen, Tabellen, Literatur, Bildquellen)S. 91
  8. Anhang A (Anlagen) und Anhang B (Selbstständigkeitserklärung)S. 101

Key findings

  1. 1

    Classic text analysis is sufficient for phrase search – neural networks only pay off there once usage data allows a personalised ranking.

  2. 2

    Neural networks are indispensable for image search because people search for image content; image titles and OCR text complement the insufficiently trained network.

  3. 3

    Combining both approaches is the right way to build a web search engine.

  4. 4

    The prototype only scales with clustering, strictly encapsulated modules and a temporary link database in front of the search database.

  5. 5

    A new search engine for Europe would have to be transparent, open source, ad-free, donation-funded and hosted in Europe – and would still hardly stand a chance against the market leaders.

All 88 references of the thesis are listed by topic on the references page. To the references

Read the thesis

The published version, served from this site (3.3 MB). No third-party connections.

Official publication

The thesis is published in the open-access repository of the University and State Library of Saxony-Anhalt (Share_it) and permanently citable via DOI and URN. Licence: Creative Commons Attribution 4.0.

How to cite

APA

Imhof, L. (2019). Information Retrieval für eine Web Suchmaschine mithilfe von neuronalen Netzen und klassischen Methoden [Bachelorarbeit, Hochschule Merseburg]. Hochschulbibliothek Merseburg. https://doi.org/10.25673/13738

BibTeX

@thesis{imhof2019zypersearch,
  author = {Imhof, Lucas},
  title  = {Information Retrieval f{\"u}r eine Web Suchmaschine mithilfe von neuronalen Netzen und klassischen Methoden},
  type   = {Bachelorarbeit},
  school = {Hochschule Merseburg},
  year   = {2019},
  month  = mar,
  doi    = {10.25673/13738},
  url    = {https://opendata.uni-halle.de/handle/1981185920/13836},
  note   = {Lizenz: CC BY 4.0}
}