Theses
Theses
Marin Kukovačec
Strojno otkrivanje sarkazma u komentarima korisnika društvenih mreža
Automated Sarcasm Detection in Social Network Users' Comments
2017
Undergraduate
Jan Šnajder
Zoran Medić
FER
FER2
5322
35
EN
Sarkazam je ironična ili satirična primjedba koja izgleda kao da nekoga hvali, ali zapravo ga ismijava. Područje računarske znanosti koje se bavi problemima poput detekcije sarkazma je obrada prirodnog jezika, koja često koristi strojno učenje kako bi polučila najbolje rezultate. U ovom završnom radu opisana je implementacija rješenja za problem detekcije sarkazma u komentarima korisnika na društvenoj mreži Facebook. Unutar problema koji se obra ̄duje nalaze se tri podzadatka: detekcija sarkazma u rečenicama izva ̄denim iz sarkastičnih komentara, detekcija sarkazma u postovima bez vanjskog konteksta, te detekcija sarkazma u postovima sa kontekstom. Svaki podzadatak je zahtijevao poseban skup podataka za treniranje modela, kao i poseban i jedinstven testni skup podataka za evaluaciju modela. U ovom radu su opisani izgled skupa podataka svakog podzadatka, sama implementacija rješenja i rezultati evaluacije nad testnim skupom podataka za svaki od podzadataka.
Sarcasm is an ironic or satirical remark that seems to be praising someone or something but is really taunting or cutting. Field of computer that deals with problems like sarcasm detection is natural language processing, which often uses machine learning to yield best results. This Bachelor thesis describes solution implementation of automated sarcasm de- tection in users comments on Facebook social network. This task of sarcasm detection can be split in three subtasks: Sarcasm detection in sentence fragments taken from sarcastic posts, Sarcasm detection in post fragments without their outside context, and Sarcasm de- tection in posts with their outside context. Each subtask required unique splitting of original dataset into train dataset to train the model, and test dataset to evaluate the model. This the- sis describes dataset of each subtask, solution implementation, and evaluation results on test datasets for each subtask.
obrada prirodnog jezika, strojno učenje, detekcija sarkazma, društvene mreže, hrvatski jezik
natural language processing, machine learning, sarcasm detection, social networks, Croatian language
6.7.2017.
Sentiment analysis from social network users' comments has a wide range of applications, including market research, customer experience analysis, political sciences, etc. A major challenge in sentiment analysis from social media texts is the abundant use of sarcasm. Sarcasm, or verbal irony, refers to the statements whose intended meaning is the opposite from the one expressed in text, and which typically serves to mock or convey discontent.
The topic of this thesis is sarcasm detection in telecom users' Facebook comments in Croatian language. Study the methods for sentiment analysis and the methods for sarcasm detection in social network users' comments. Devise and implement a method for sarcasm detection based on machine learning and in-context contrastive sentiment analysis. Compile a suitable dataset for model training and evaluation. Carry out an experimental evaluation of the method on the test set, including a detailed error analysis. All references must be cited, and all source code, documentation, executables, and datasets must be provided with the thesis.