Predicting Word-Guessing Times in the Wordle Game using GBDT and FCNN

With the popularity of Wordle Game in recent years, more and more people have been studying the complex word-guessing rules. Majority of researchers focus on classifying the difficulty of word guessing, thus neglecting the study of word guessing time. This paper proposes to get a model that can predict the distribution of guessing times for given words. Firstly, we use the two specified attributes obtained by conducting a preliminary analysis to train the number of correct guesses for each category, and each category will be input to the network seven times to obtain the final result. Afterwards, we proceed to train both Gradient Boosting Decision Tree(GBDT) prediction model and Fully Connected Neural Network(FCNN) model. In the end, we obtain the prediction results of guessing times for given words. The results show that the prediction accuracy of our model is 95%. It illustrates that the model has great performance which is suitable for most applications.