prikaz prve stranice dokumenta Clickbait Titles Detection with Machine and Deep Learning Methods
  Download
PDF 7.42 MB
master's thesis
Clickbait Titles Detection with Machine and Deep Learning Methods
2023. urn:nbn:hr:195:568022
Krivičić, Armin
University of Rijeka
Faculty of Informatics and Digital Technologies

Cite this document

Krivičić, A. (2023). Clickbait Titles Detection with Machine and Deep Learning Methods (Master's thesis). Rijeka: University of Rijeka. Retrieved from https://urn.nsk.hr/urn:nbn:hr:195:568022

Krivičić, Armin. "Clickbait Titles Detection with Machine and Deep Learning Methods." Master's thesis, University of Rijeka, 2023. https://urn.nsk.hr/urn:nbn:hr:195:568022

Krivičić, Armin. "Clickbait Titles Detection with Machine and Deep Learning Methods." Master's thesis, University of Rijeka, 2023. https://urn.nsk.hr/urn:nbn:hr:195:568022

Krivičić, A. (2023). 'Clickbait Titles Detection with Machine and Deep Learning Methods', Master's thesis, University of Rijeka, accessed 05 May 2024, https://urn.nsk.hr/urn:nbn:hr:195:568022

Krivičić A. Clickbait Titles Detection with Machine and Deep Learning Methods [Master's thesis]. Rijeka: University of Rijeka; 2023 [cited 2024 May 05] Available at: https://urn.nsk.hr/urn:nbn:hr:195:568022

A. Krivičić, "Clickbait Titles Detection with Machine and Deep Learning Methods", Master's thesis, University of Rijeka, Rijeka, 2023. Available at: https://urn.nsk.hr/urn:nbn:hr:195:568022

Please login to the repository to save this object to your list.
Metadata
TitleClickbait Titles Detection with Machine and Deep Learning Methods
Title (croatian)Otkrivanje “clickbaitova” pomoću metoda strojnog i dubokog učenja
AuthorArmin Krivičić
MentorSanda Martinčić-Ipšić (mentor)
Committee memberMiran Pobar (predsjednik povjerenstva)
Committee memberSlobodan Beliga (član povjerenstva)
GranterUniversity of Rijeka
(Faculty of Informatics and Digital Technologies)
Rijeka
Defense date and country2023-09-26, Croatia
Scientific / art field,
discipline and subdiscipline		SOCIAL SCIENCES
Information and Communication Sciences
Abstract
This master's thesis addresses the challenge of clickbait detection in digital media through the application of machine learning and deep learning techniques. Clickbait is a type of web content advertisement that aims to attract users' attention and encourage them to click on related links. Media scholars have expressed concerns regarding clickbait's contribution to misinformation, yet the clickbait industry is still growing quickly. The objective of this thesis is to classify clickbait titles using different models and text representation techniques. Multiple datasets, including the dataset collected by Chakraborty et al. and the Webis-Clickbait-17 dataset, are utilized in the research. Furthermore, these two datasets were merged, and classification was also done on that merged dataset. Classification models such as the naive Bayes classifier, support vector machines, random forest, linear perceptron, LSTM, and GRU are employed. Three text representation techniques, TF-IDF, word2vec, and FastText, are used and compared. The findings demonstrate that FastText representation yields the highest overall performance. Regarding the classification performance, support vector machines achieve the best results, with an F1 score of 0.8797 on the merged dataset. The study also explores title-text similarity using the neural Deep Semantic Similarity Model, confirming that non-clickbait titles better represent the content. This research contributes insights into clickbait classification models, aiming to enhance user experience and combat misinformation.
Abstract (croatian)
Tema diplomskog rada je otkrivanje "clickbaitova" (mamilica) u digitalnim medijima koristeći tehnike strojnog i dubokog učenja. "Clickbait" je vrsta oglašavanja web sadržaja kojoj je cilj privući pozornost čitatelja i potaknuti ih da kliknu na popratne poveznice. Industrija "clickbaitova" i dalje ubrzano raste bez obzira na zabrinutost vezanu uz širenje dezinformacija koju su iskazali znanstvenici. Cilj rada je klasificirati "clickbait" naslove koristeći različite modele i reprezentacije teksta. Više skupova podataka je korišteno u istraživanju, uključujući skup podataka koji su prikupili Chakraborty et al. i skup podataka Webis-Clickbait-17. Nadalje, ova dva skupa podataka spojena su i klasifikacija je također urađena na takvom skupu podataka. Koriste se modeli kao što su naivni Bayesov klasifikator, strojevi s potpornim vektorima, slučajna šuma, linearni perceptron, ćelija s dugoročnom memorijom (LSTM) i upravljačka rekurentna jedinica (GRU). Korištene su i uspoređene tri tehnike reprezentacije teksta, TF-IDF, word2vec i FastText. Rezultati pokazuju da FastText reprezentacija ima najbolje performanse. Što se tiče klasifikacijskih performansi, strojevi s potpornim vektorima postižu najbolje rezultate klasifikacije, s F1-rezultatom od 0.8797 na spojenom skupu podataka. Studija također istražuje sličnost naslova i tekstova članaka i pokazalo se da naslovi koji nisu "clickbaitovi" bolje predstavljaju sadržaj. Ovo istraživanje pruža uvid u modele klasifikacije "clickbaitova", a s ciljem poboljšanja korisničkog iskustva i borbe protiv dezinformacija.
Keywords
Keywords (croatian)
Languageenglish
URN:NBNurn:nbn:hr:195:568022
Study programmeTitle: Informatics
Study programme type: university
Study level: graduate
Academic / professional title: magistar/magistra informatike (magistar/magistra informatike)
Type of resourceText
File originBorn digital
Access conditionsOpen access
Terms of useLicence Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) icon
Created on2023-09-21 18:57:13