採用隱含狄利克雷分佈分析使用者評論：以Steam網站為例

Abstract

過去的研究大多著重於檢驗電子遊戲專家評論，僅有少數研究在探索遊戲玩家提供的使用者生成內容(UGC)。使用者生成內容包含玩家感受的豐富資訊，有助於遊戲開發商了解玩家的真實想法。基於使用者生成內容，本研究著重於分析玩過角色扮演遊戲(Role-Playing Game，簡稱RPG)玩家的使用者生成內容，以探索他們的想法和態度。本研究提出一個適用於遊戲評論文本主題的分析方法，主要是依據隱含狄利克雷分布(Latent Dirichlet Allocation，簡稱LDA)演算法，它可以將文本集中成一篇，而每篇文本的主題則按照概率分布的形式。從Steam遊戲平台所擷取的玩家評論內容中或許含有關於角色、遊戲畫面、故事等相關的評論主題，這些主題可以視為玩家的特徵，因此本研究運用LDA主題模型，從玩家的遊戲評論內容分析出玩家特徵以及遊戲的關鍵屬性。

綜合研究結果可以確認，本研究選擇的三款目標遊戲雖都同屬於角色扮演遊戲類型，但遊戲內容及玩法不盡相同(如故事背景、遊戲視角等)。本研究將評論資料基於LDA主題模型分析出大部分玩家著重於對於角色扮演遊戲的規則及目標，但由於三款遊戲還是有不同內容，因此從LDA主題模型分析出也會有相異的主題屬性。

本研究提供遊戲開發商一個實務方式來有效率確認玩家關注的主題，對於遊戲的後續改善提出建議，以及開發新遊戲提出方向。In the past, most of the studies focused on game reviews from experts, with only a few exploring the reviews provided by players. User reviews contain a wealth of information about players' feelings and helps developers understand players’ thoughts. Based on user-generated content, this study focuses on analyzing the user-generated content of role-playing games players to explore their thoughts and attitudes.

This study proposes an analysis method suitable for game reviews, mainly based on the Latent Dirichlet Allocation(LDA), which can collect all the reviews into one document, and the topic of each is in the form of probability distribution given. The player review contents extracted from Steam website may contain some topics such as characters, graphics, stories, etc. These topics can be considered as the characteristics of the player. Therefore, this study uses the LDA topic model to analyze the player's reviews and analyzes players' characteristics and key attributes of the games.

Based on the results, it can be confirmed that although the three different games selected are role-playing games, however the game content and gameplay are different (such as story background, game perspective, etc.). This study analyzes the review based on the LDA topic model and finds that most players focus on the rules and objectives of role-playing games, but since these three games still have different contents, the LDA topic model will also have different topic attributes.This study provides an empirical method for developers to identify the topics that players focus on and gives suggestions for the following improvement.