Home |  ENGLISH |  Kontakt |  Impressum |  Datenschutz |  Anmelden |  KIT

Thema4380: Unterschied zwischen den Versionen

Aus Aifbportal

Wechseln zu: Navigation, Suche
 
Zeile 4: Zeile 4:
 
|Nachname=Bie
 
|Nachname=Bie
 
|Abschlussarbeitstyp=Master
 
|Abschlussarbeitstyp=Master
|Betreuer=Harald Sack; Rima Türker
+
|Betreuer=Ralf Reussner; Harald Sack; Rima Türker
 
|Partner=FIZ Karlsruhe;
 
|Partner=FIZ Karlsruhe;
 
|Forschungsgruppe=Information Service Engineering
 
|Forschungsgruppe=Information Service Engineering

Aktuelle Version vom 10. Oktober 2019, 09:21 Uhr



Words Are Not enough! Short Text Classification using Words and Entities


Qingyuan Bie



Informationen zur Arbeit

Abschlussarbeitstyp: Master
Betreuer: Ralf ReussnerHarald SackRima Türker
Forschungsgruppe: Information Service Engineering
Partner: FIZ Karlsruhe
Archivierungsnummer: 4380
Abschlussarbeitsstatus: Abgeschlossen
Beginn: 25. Oktober 2018
Abgabe: 09. Oktober 2019

Weitere Informationen

Text classification is gaining more attention due to the availability of huge numbers of text data, which includes search snippets, news data as well as text data generated in social networks. Recently, several supervised learning approaches have been proposed for text classification. Most of them use words to generate a feature set and adopt machine learning algorithms. However, if applied to short texts, most of the standard text classification approaches suffer from issues such as data sparsity, and insufficient text length. Moreover, due to the lack of contextual information, short texts are highly ambiguous. As a result, simple text classification approaches based on words only, cannot represent the critical features of short texts properly. Thus, short text classification is much more challenging in comparison to traditional long documents.

In this thesis, to overcome the mentioned shortness and sparsity problem of short text, we will enrich the text representation by leveraging words together with entities represented by the content of the given document. More specifically, the feature set of a given (short) text will be composed of words and entities of the text. To extract the entities present in a text, existing Entity Linking Systems can be applied, such as TagMe[1].

The aim of the thesis is to develop a supervised classification based approach to classify a given short document. In the first step, existing embedding approaches such as Word2Vec[2] will be used in order to map each word and entity to a multidimensional vector space. Next, by utilizing the embeddings the feature set for the subsequent text classification will be generated. Finally, existing Machine Learning and/or Deep Learning methods will be applied to complete the classification task.

This thesis will be supervised by Prof. Dr. Harald Sack, Information Service Engineering at Institute AIFB, KIT, in collaboration with FIZ Karlsruhe. [1] https://tagme.d4science.org/tagme/ [2] https://goo.gl/vGusdA


Ausschreibung: Download (pdf)