Data Mining Study of Wordle Based on Multi-Model Prediction

As a game that spreads across the Internet and stimulates people to think and learn, data mining of Wordle's report counts and their distribution is particularly important. In this paper, an ARIMA model and a BP neural network model based on the WOA are developed to perform in-depth data mining on Wordle. For model I, this paper uses ARIMA model fitting to fit the trend of the total number of historical report results over time, and obtains the R2 of the model as 0.986. Based on this, this paper establishes a prediction model for the interval of the total number of reports, and gives the prediction interval of the total number of report results on March 1, 2023 as [21211], [21814]. For Model II, this paper defines word attributes as the following four indicators: for example, the commonness of the word, the number of occurrences of the same letter in the word. The Spearman correlation coefficient between the percentage of reported scores to the total reported scores in the Hard mode and the word attributes was calculated. Finally, in the Hard model, the percentage of reported scores out of the total reported scores was significantly correlated with how common the word was and whether it began with a vowel letter. For Model III, this paper constructed a BP Neural network model based on WOA, which, in this paper, uses word attributes as the input layer and the report distribution of words as the output layer. In the WOA, the optimal weights and thresholds of the BP neural network are obtained in this paper by continuous search and approximation. The final model has a Mean Absolute Error value of 0.38, both of which have good performance. At the same time, this paper predicts the results of “EERIE”, and the obtained distribution results are rounded to [1], [7], [20], [22], [21], [19], [10]. Through the above modelling and calculation, this paper mines Wordle's user data, which has some theoretical significance for the healthy development of Wordle.