First, an ARIMA-based prediction model for the number of reported outcomes at future dates was developed. After processing the collected data, the smoothness of the number of reported outcomes was verified, and an intrinsic trend analysis of the data was performed to demonstrate the validity of the ARIMA model for predicting the number of reported outcomes at future dates, and the parameters of ARIMA were fitted using historical data to determine the final prediction model as ARIMA. secondly, one prediction model based on the GA-BP neural network algorithm was developed for the relevant percentages of future dates. At the same time, the GA algorithm was introduced for optimization to obtain the optimal initialization of the BP neural network. We use the frequency of letter occurrences together with the date as the input word feature and the correlation percentage as the output, and the prediction results are within the valid interval. Third, the word difficulty classification model and word scoring model were built based on the improved k-means clustering algorithm and PCA, respectively. To improve the drawback that the k- value in k-means algorithm cannot be determined, were introduced CH value, DB value. as indicators to evaluate the k- value, and combined the entropy weight method with Topsis to get the optimal number of clusters as 3 and the clustering index as the correlation percentage, and then applied the k-means algorithm to classify the words. We used the word difficulty ratio to determine the difficulty coefficient of each category, and also combined the score interval of each category obtained by PCA. Fourth, we found that the Wordle trivia game has the characteristic of viral spread in blocks, and more and more players choose the difficult mode as time goes by.