Prediction of the Reported Results of Wordle Based on ARIMA and GBDT

Wordle is a simple but addictive scrabble game run by The New York Times, in which players have to guess a five-letter solution word in six attempts. Many people share their experiences and today's scores on Twitter. In this paper, the results shared by players on Twitter from January 7, 2022 to December 31, 2022 are crawled. ARMA(5,6) and ARMA(5,7) are constructed based on time sequence analysis and heat map to analyze the change of the number of reported results over time. And a forecast interval of the number of reported results on March 1, 2023 is given. Furthermore, this study adopts the Grand causality test to analyze the relationship between the word attributes, the frequency of usage, and the percentage of reported scores in Hard Mode. It was found that the attributes of words 1 and 2 days ago would have an impact on the percentage data of that day. This paper also constructs the Gradient Boosting decision tree model and predicts the distribution of results corresponding to a given date and a given solution word.