Результаты исследований: Публикации в книгах, отчётах, сборниках, трудах конференций › статья в сборнике материалов конференции › научная › Рецензирование
Topic Modelling with NMF vs. Expert Topic Annotation : The Case Study of Russian Fiction. / Sherstinova, Tatiana; Mitrofanova, Olga; Skrebtsova, Tatiana; Zamiraylova, Ekaterina; Kirina, Margarita.
Advances in Computational Intelligence - 19th Mexican International Conference on Artificial Intelligence, MICAI 2020, Proceedings. ред. / Lourdes Martínez-Villaseñor; Hiram Ponce; Oscar Herrera-Alcántara; Félix A. Castro-Espinoza. Springer Nature, 2020. стр. 134-151 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Том 12469 LNAI).Результаты исследований: Публикации в книгах, отчётах, сборниках, трудах конференций › статья в сборнике материалов конференции › научная › Рецензирование
}
TY - GEN
T1 - Topic Modelling with NMF vs. Expert Topic Annotation
T2 - 19th Mexican International Conference on Artificial Intelligence, MICAI 2020
AU - Sherstinova, Tatiana
AU - Mitrofanova, Olga
AU - Skrebtsova, Tatiana
AU - Zamiraylova, Ekaterina
AU - Kirina, Margarita
N1 - Publisher Copyright: © 2020, Springer Nature Switzerland AG.
PY - 2020
Y1 - 2020
N2 - The paper presents an experiment aimed at comparison of results of topic modelling via non-negative matrix factorization (NMF) with that of manual topic annotation performed by an expert. The experiment was conducted on the annotated corpus of Russian short stories of the initial three decades of the 20th century, which contains 310 stories with a total of 1000000 tokens written by 300 Russian writers. The annotation scheme used in topic annotation includes 89 topics, further this list was reduced down to 30 generalized ones, the most frequent of which turned out to be the following: death, relationships, love, social groups, social processes, family, money, human sins, nature, religion, and war. Then, the corpus divided into three consecutive time periods was subjected to NMF topic modelling which provided a model including 24 topics. The results of both topic annotations were compared and described. The paper discusses the main findings of the study and the difficulties of fiction topic modelling which should be taken into account. For example, experimental results showed that topic modelling via NMF should be primarily recommended for the revealing of topics referring to general background of literary texts (e.g., war, love, nature, family) rather than for detecting topics related with some critical events or relations between characters (e.g., death or relations). The comparison of human and automatic topic annotation seems an important step for the improvement of artificial technologies techniques related with NLP.
AB - The paper presents an experiment aimed at comparison of results of topic modelling via non-negative matrix factorization (NMF) with that of manual topic annotation performed by an expert. The experiment was conducted on the annotated corpus of Russian short stories of the initial three decades of the 20th century, which contains 310 stories with a total of 1000000 tokens written by 300 Russian writers. The annotation scheme used in topic annotation includes 89 topics, further this list was reduced down to 30 generalized ones, the most frequent of which turned out to be the following: death, relationships, love, social groups, social processes, family, money, human sins, nature, religion, and war. Then, the corpus divided into three consecutive time periods was subjected to NMF topic modelling which provided a model including 24 topics. The results of both topic annotations were compared and described. The paper discusses the main findings of the study and the difficulties of fiction topic modelling which should be taken into account. For example, experimental results showed that topic modelling via NMF should be primarily recommended for the revealing of topics referring to general background of literary texts (e.g., war, love, nature, family) rather than for detecting topics related with some critical events or relations between characters (e.g., death or relations). The comparison of human and automatic topic annotation seems an important step for the improvement of artificial technologies techniques related with NLP.
KW - Corpus linguistics
KW - Digital humanities
KW - Fiction
KW - Literary criticism
KW - Machine learning
KW - NMF
KW - NPL
KW - Russian literature
KW - Topic modelling
UR - http://www.scopus.com/inward/record.url?scp=85092935662&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-60887-3_13
DO - 10.1007/978-3-030-60887-3_13
M3 - Conference contribution
AN - SCOPUS:85092935662
SN - 9783030608866
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 134
EP - 151
BT - Advances in Computational Intelligence - 19th Mexican International Conference on Artificial Intelligence, MICAI 2020, Proceedings
A2 - Martínez-Villaseñor, Lourdes
A2 - Ponce, Hiram
A2 - Herrera-Alcántara, Oscar
A2 - Castro-Espinoza, Félix A.
PB - Springer Nature
Y2 - 12 October 2020 through 17 October 2020
ER -
ID: 98682414