Wordle result prediction based on machine learning algorithm LSTM model

Wordle, a novel word puzzle courtesy of the New York Times, is becoming popular on the Internet. Since players are willing to share their scores on Twitter, this adds a social attribute to the game. We first conducted a multi-factor ANOVA on the attributes of the words and the percentage of scores in the difficult mode. The results showed that the letters in the first four positions of the word and the number of repeated letters in the word had an effect on the percentage of scores. Next, we proposed a multiple-input-multiple-output long and short-term memory (LSTM) model to predict the distribution of reported outcomes. To incorporate the internal effects of the data into the study, we designed a six-input-seven-output LSTM regression prediction model and trained it with the dataset. Then, the trained network was used to predict the distribution of "EERIE" scores. To predict the difficulty of words, a random forest model was designed to determine which of these attributes were associated with each classification by using several attributes carefully selected from the words. Finally, the difficulty of the word "EERIE" was predicted.