Analysis of Wordle Game Mechanism Based on LightGBM Algorithm and Word Feature Analysis

The popularity of Wordle has attracted a large number of players to report their scores on Twitter. Today, Wordle has been acquired by the New York Times. This paper makes some data analysis on how to maintain the enthusiasm of players and play the positive role of the game more effectively. The frequency of the letters in each position of the word, the number of repeated letters and the number of repeated letters in the word are extracted, and the common use rate of each word is calculated based on the data of Google Ngram Viewer. Qualitative analysis is made on the percentage distribution of the reported score played in difficult mode through attributes such as thermal diagram and skewness, and the conclusion is drawn that the number of repeated letters in the word and the number of repeated letters have a negative correlation with the score played, while the common use rate of words and the frequency of letters in each position of the word have a positive correlation with the score played. On the basis of fully considering the influence of the word attribute, we select the date, the common use rate of the word and the frequency of the letters in the five positions of the word as the characteristic indicators that affect the distribution of the results, and establish the LightGBM model for the percentage of (1, 2, 3, 4, 5, 6, X). Finally, the trained seven models are used to analyze and predict the distribution of the percentage of the times used to guess the words in the future. The R2 coefficient of the model in the training set is 0.85, and the R2 coefficient in the test set is 0.68, indicating that the model has good fitting and prediction ability. We use the LightGBM model to predict the percentage of times the word ERRIE will be guessed on March 1, 2023, and the distribution is [0.129, 2.152, 13.861, 36.167, 28.278, 14.002, 5.411].