Research on Word Attributes and Word Meaning Difficulty Based on Time Series Analysis

Abstract

In recent years, natural language processing has increasingly focused on revealing the relationship between word attributes and word meaning difficulty through data mining and statistical modeling to provide new perspectives for lexical cognition research. In this study, the daily result data from January 7 to December 31, 2022, classify word attributes into 30 categories of related attributes. A time-series model was built to derive the number of reported results on March 1, 2023, to be about 12,295 items. It was found that 29 attributes such as the commonness of words are not related to the difficulty pattern and the meaning of words are related to the difficulty pattern. Then corresponding to the data on the number of attempts to solve different subcategories, a gradient boosting tree (GBDT) [1] regression model was built to derive the correlation percentages under the number of attempts 1, 2, 3, 4, 5, 6, and X for players solving the EERIE word case as 1%, 4%, 18%, 28%, 28%, 16%, and 5%. In this paper, it is found that in solving EERIE words, players have a higher probability of solving the word under 4 and 5 attempts and a smaller probability of solving the word under 1 and 6 attempts.