A black-box model for predicting difficulty of word puzzle games: a case study of Wordle.

Autor: Shi, Ling, Chen, Yingke, Lin, Jiaxuan, Chen, Xiaoyu, Dai, Guangming
Předmět:
Zdroj: Knowledge & Information Systems; Mar2024, Vol. 66 Issue 3, p1729-1750, 22p
Abstrakt: The popular word-filling game Wordle has gained widespread attention since its release in 2022. Much attention has been paid to find the optimal strategy. However, this article proposes a black-box prediction model that can accurately predict the difficulty level of words in the game to find the deep rules in the game data. In this work, we scientifically established a black-box model for game difficulty prediction. We achieve high accuracy in new datasets and show strong stability in similar tasks. The black-box model is divided into the game input content feature extraction model and the game output content rule extraction model. This research scientifically and effectively extracts word attributes, including word frequency, letter frequency, part of speech, times of letter repetitions, and word meaning score from the input content. Then it reduces the seven kinds of proportion of people in different tries in output content into two indices using the Critic method. Finally, it establishes a gradient boosting decision tree-based multiple regression model, making the final prediction accuracy of difficulty level for new words reach 95%. It is believed that the black-box prediction model can provide valuable insights for game designers and developers. And the research provides an innovative method to predict and understand user behavior in online games, contributing to the broader field of data science. The integration of data-driven methodologies in the gaming industry opens new possibilities for understanding player interactions and further enhancing game development strategies. [ABSTRACT FROM AUTHOR]
Databáze: Complementary Index
Externí odkaz: