The Application of Data Analysis in A Web-Based Word Game: Worldle

Abstract

We will analyze the game Worldle based on its data in 2022. First, we use the EEMD Model to analyze the trend of the number of players over time in this year.We need to forecast the number of players in the next 60 days and reduce the prediction error, we use the Moving Average and build Rolling Window to reduce the forecast error. We use both LSTM model and EEMD-LSTM for training and comparing the effects of the two and find EEMD-based LSTM model is better. Then, we come up with several properties of the solution words that we think would affect the difficulty of the puzzle, and after Correlation Analysis, we find that this is indeed the case and give some explanations. Second, we mainly use Deep Learning. To predict the associated percentages of (1, 2, 3, 4, 5, 6, X) for a given future solution word on a future date, we use a Genetic Algorithm and Back Propagation Neural Network (GA-BP Neural Network) to analyze the percentage distribution of given words. After that, the distribution of the number of answers for the example can be obtained by prediction as (1.1831,6.2839,21.3307,28.6500,24.7627,14.7354,3.8857). Third, how difficult is it to guess a word correctly? For this problem, we use a method belonging to Machine Learning - PSO Decision Trees. Before this, our group first used RSR and the Pearson Correlation Coefficient to divide the difficulty level of words.