Research on Scrabble Prediction Based on Neural Network Model

Recently, Wordle is a daily Scrabble game provided by the New York Times, which has been loved by many players. Based on this game, this paper focuses on predicting the Scrabble game based on neural network models, using methods such as One-Hot Encoding, correlation test, and so on to solve the game, and the specific task is divided into four major problems. For problem one, firstly, the given data are preprocessed, some outliers are eliminated, while the original data are replaced by the mean value of the number of people, and the filtered data are normalized to the range [0, 1]. Then, a Nested Two-Layer LSTM model is constructed to predict the number of reported results, and the One-Hot Encoding property of words is analyzed to create a mapping of these discrete feature data using the OHE Finally, the correlation between the attributes and the percentage of difficulties was tested. The results showed that the number of results reported on March 1 was approximately 20520 and that the One-Hot Encoding attribute of words did not affect the percentage of scores played in the difficult mode. For problem two, a neural network model was constructed to implement multiple inputs and outputs and predict the distribution of reported results based on the feature data generated using unique heat coding in problem one. The model built uses a hidden layer with 10 neurons. Meanwhile, the model was used to predict the word EERIE, and the results showed that the distribution of the results for the word EERIE was [0.5,7.1,23.9,30.5,23.6,11.8,2.0], and the average absolute percentage error of the model on the test set reached 9%, which proved that the model had a good fitting ability. For problem three, the difficulty was first set to three levels of difficulty, medium and easy, and the classification criteria were judged based on the percentage of three or fewer passes, a convolutional neural network model was constructed using deep learning techniques. The results showed that the model constructed using convolutional neural networks predicted the best results and yielded a moderate difficulty level for the word EERIE.