An analytical study of Wordle development based on multimodality

In this paper, a BP neural network prediction model optimised based on genetic algorithm and a difficulty-based classification model are developed for predicting and classifying the metrics for a given word on a future date. In this study, the genetic algorithm optimised BP neural network was first used to analyse the prediction of the number of results to be reported on 1 March. The results show that the prediction interval for the number of results reported on 1 March is (21021, 22545), and the average absolute percentage error for the test set is only 3.5%, which is a good prediction. Next, the paper identifies the word attributes that influence the percentage of times required to guess a word in the difficult mode, listing four word attributes that may have an impact. In this paper, we use Spearman's correlation analysis to analyse the relationship between each word attribute and the percentage of scores. In order to simplify the model, a word difficulty evaluation model was established using the TOPSIS algorithm based on the entropy weighting method to objectively calculate the word difficulty score. Finally, it was concluded that the word attributes that affect the score percentage are: the number of commonly used letters, the word usage level and the number of letter repetitions. In order to classify word difficulty, on the basis of word difficulty score, word difficulty was classified into four categories according to the Euclidean distance between elements through K-means clustering analysis. Then, the percentage of each score of the word to be classified was derived through the regression prediction model, and the derived percentage was substituted into the evaluation model in Task 1 to obtain the difficulty score of the word and finally determine the category to which it belongs. The results show that the difficulty score of EERIE is 0.409, which belongs to the more difficult category.