Puzzle Game Data Analysis Based on Big Data and Neural Network Algorithms

Wordle is a popular daily puzzle game offered by The New York Times, challenging players to guess a five-letter word with iterative feedback. This game has amassed a substantial Twitter following, providing a valuable dataset for analysis. In this study, we leverage a dataset of 10,000 tweets spanning two months to investigate the trends and patterns of Wordle players. We introduce two neural network models to forecast the quantity and distribution of future reported results. The first model is founded on Long Short-Term Memory (LSTM) recurrent neural networks, utilizing the 10-day rolling mean as the primary prediction sequence. The second model employs a Genetic Algorithm-optimized Backpropagation (GA-BP) temporal network to fine-tune the weights and thresholds of the Backpropagation neural network. We benchmark these models against established methods such as ARIMA and linear regression. Our findings reveal that the LSTM recurrent neural network achieves the highest accuracy in forecasting the number of reported results, exhibiting an impressively low Root Mean Square Error (RMSE) value of 1.794. For forecasting the distribution of reported results, the GA-BP temporal network excels with an exceptional R-squared (R-value) of 0.99752. This study underscores the potential and utility of neural network models in the analysis of social media data related to online games. Furthermore, it offers valuable insights into the behaviors and preferences of Wordle players, thereby informing the design and enhancement of similar games in the future.