Predictive Analysis Based on GBDT Deep Learning Regression Model in Wordle Games

Wordle is a popular puzzle currently offered daily by the New York Times. At present, there are daily game result files of users who have participated in this game for nearly one year. The document includes the date; Competition number; Words of the day; Number of people reporting scores on that day; Number of players entering difficult mode; The percentage of words that can't be solved after 1~6 attempts. Based on this file, this paper mainly studies the influence of word attributes on report score, and creates a model to predict the distribution of report results. The attributes of words are divided into six influencing factors (Fre, Count, Sum, Repeat, Vowel, Root), and a polynomial fitting model is established to get the specific relationship. Then an ARIMA time series model is established to verify the accuracy of the model and analyze the errors. Then, according to the six influencing factors (inputs) and the relevant percentages (outputs) of future dates (1,2,3,4,5,6,X) in the question, the GBDT deep learning regression model of gradient lifting tree is established. So as to study the influence of word attributes on the percentage of relevance, and finally determine the result.