期刊名称:BIBLOS : Revista do Instituto de Ciências Humanas e da Informação
印刷版ISSN:0102-4388
电子版ISSN:2236-7594
出版年度:2020
卷号:34
期号:2
页码:344-365
DOI:10.14295/biblos.v34i2.12395
出版社:Universidade Federal do Rio Grande
摘要:This research's general objective was to analyze if there is a temporal variation characteristic of the distribution of values of relevant terms over the time of the production of texts that can contribute as a criterion for the automatic indexing process. The doctoral theses of the graduate programs (PPGs) in the area of Applied Social Sciences at UFMG were analyzed, considering seven different PPGs, each of which is a corpus, with 641 theses defended in a period of twelve years, from 2007 to 2018. The terms considered were all the noun phrases contained in the texts of the theses. Each noun phrase received a value associated with its relevance as a descriptor according to the term frequency criteria in the thesis itself (TF - Term Frequency) and with the inverse of the frequency of occurrence of the term in the total of theses of each PPG (IDF - Inverse Document Frequency). The theses were divided into 12 groups in each PPG to calculate the average defense date of the theses and the average consolidated score of the relevant terms in the theses. As a result, each PPG's characteristic behavior was identified through a scatter plot of the average level of relevance score over time. For each graph of each of the 7 PPGs, a trend line was added, considering its respective R², and its specific analysis was made. All temporal distribution behaviors were characterized in polynomial equations and applied as a criterion for automatic indexing.
关键词:Recuperação da Informação Temporal;Indexação Automática;Sintagmas Nominais