Data analysis based on word difficulty affecting the number of users in Wordle games

Wordle is an English spelling game where players have to guess a word with five English letters six times in a day. This paper first preprocesses the raw data of the game, builds a prediction model to predict the result on March 1 as 10456 through the results of the ARIMA model, and verifies that the properties of the words do not affect the percentage of scores in the reported hard mode through the combination of Pearson correlation analysis and linear regression. Secondly, the distribution of n success words is predicted by establishing a multilayer perceptron (MLP) and linear regression model based on the regression chain. Next, we used the TOPSIS method to grade the word difficulty, and finally selected the convolutional neural network (CNN) model. Finally, after some exploration, we found that 3 times were highly negatively correlated with 5 and 6 times.