Naslov
Classification of text according to emotions using matrix factorizations

Autori
Dobša, Jasminka ; Bužić, Dalibor

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
Book of Abstracts of the ISCCRO'20 - International Statistical Conference in Croatia / Berislav Žmuk, Anita Čeh Časni - Zagreb : Hrvatsko statističko društvo, 2020, 13-13

Skup
3rd International Statistical Conference in Croatia (ISCCRO'20)

Mjesto i datum
Zagreb, Hrvatska, 15.10.2020. - 16.10.2020

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
classification of textual documents, LSI, Reduced k-means, sentiment analysis

Sažetak
Sentiment analysis is one of the fast growing research area in the field of text mining. It includes two major approaches: the first is supervised, based on annotated data sets, and the second is unsupervised, based on lexicons. In this research a few models as extension of vector space model are proposed and compared. For that purpose two matrix factorizations are used: the first is singular value decomposition and the second is factorization by usage of Reduce kmeans method. Representation of documents in vector space model using truncated singular value decomposition is known in text mining as method of Latent Semantic Indexing. By usage of method of Reduced k-means for factorization of term-document matrix the clustering structure of the data set should be preserved which enables representation of documents in a form suitable for classification. Further on, usage of Reduced k- means method enables hybrid approaches by inclusion of lexicons. Experiments are conducted by usage of ISEAR data set where sentences are labelled by emotions (joy, fear, anger, sadness, disgust, shame and guilt). In further research, it is planned to propose hybrid methods by representation of words by global matrix factorizations and deep learning approaches where words are represented by local context window methods.

Izvorni jezik
Engleski

Znanstvena područja
Informacijske i komunikacijske znanosti

POVEZANOST RADA