A study of Wordle crossword data based on machine learning algorithms

Wordle is a popular puzzle, and The New York Times provides a daily puzzle and access to user data via Twitter. We studied this document and came to some conclusions. On the one hand, we used an autoregressive integrated moving average model to predict the number of reported outcomes and the number of difficult patterns over the next 60 days. Then, we verified the correlation between word attributes and the number of people in the difficult mode. The results showed that the letters I, A, E, S and O had the strongest correlation with the number of people in the difficult mode. To assess the accuracy of the model, we changed the difference order to verify the stability of the time series data, and obtained the goodness-of-fit of the model with 0.982 and 0.948, respectively, with high accuracy. On the other hand, to classify the difficulty of words, we first implemented a semi-supervised machine learning algorithm by K-Means clustering to label words with a distribution of different attempts. The elbow rule was applied to determine the optimal number of clusters as 3. The attributes and labels of these words were used to train the KNN algorithm and to classify the difficulty of the words according to the trained model. After comparison, the accuracy of the existing model was 63%.