Word data prediction based on statistical method

Abstract

Puzzle game wordle has been wildly popular since its launch, its simple and unique game rules have generated a lot of data while gaining a large number of users in a short period of time, making wordle prediction a hot topic. In this paper, we construct four models based on the properties of words: interpretive prediction report counting model, correlation judgment model, percentage prediction model and word difficulty category classification model to provide a solution for predicting words. In order to measure the difficulty of words, we evaluated the words from two aspects of commonality and structure, and defined sub-indicators such as the harmony of word pronunciation (both American and British), the number of vowels, the number of repeated letters, and the word frequency of words to measure the difficulty of words comprehensively. We built a BP neural network model for multivalue prediction and predicted for a given future puzzle word its correlation to the percentage distribution of each attempt (1, 2, 3, 4, 5, 6, X). We developed a K-means clustering model to classify the difficulty of words. We found that words fall into two categories. The closer a word is to the center of the category; the word belongs to that category. We suggest that in addition to players taking the initiative to advertise and attract traffic, newspapers should also take the initiative to promote the game (such as advertising), firstly warming up the publicity before the game goes live. According to the popularity trend curve fitted by the differential equation model, mass interest in emerging games rises rapidly and, after peaking for a short time, slowly declines due to the loss of novelty. Therefore, in order to maintain public enthusiasm as much as possible, publicity should be stepped up after the peak.