基於文字探勘之行動遊戲推薦系統的研究

A Study of the Mobile Games Recommendation Systems Based-on Text Mining

孫慈睿(Tzu-Jui Sun)

指導教授 ： 蕭瑞祥

共同指導教授 ： 戴敏育

淡江大學/商管學院/資訊管理學系碩士班/碩士(2016年)

https://doi.org/10.6846/TKU.2016.00016

摘要

手機遊戲已成為民眾休閒娛樂中不可或缺的一部分，許多人下載遊戲時可能會參考Google Play或App Store上的評論。但在這些評論中，有些遊戲為了衝高排行，可能會用假評論來灌水。而遊戲虛擬社群-巴哈姆特遊戲資訊討論區為網友分享心得的平台，非專家或遊戲公司所編寫。本研究抓取巴哈姆特的文章資料，找出含有評價的文章並給予分數，最後將遊戲排行做推薦。
本研究採用Nunamaker(1991)等人所提出之系統發展研究法，旨在分析網路評論，整理出網友對於遊戲的評價，最後算出遊戲排名做推薦，讓使用者更容易瞭解這款遊戲的真正使用評價，且使用者的聲音也可以傳達給遊戲開發者，作為改進的考量。
本研究的實驗對象為「七騎士」、「列王的紛爭(Clash of Kings)」、「神魔之塔」三個Google Play營收排行榜前六名的遊戲，實驗資料集來自巴哈姆特的文章資料。文章蒐集範圍以討論區的子板塊「全部主題」、「心情雜談」。將含有評價的文章給予分數並人工驗證，三款遊戲準確率分別為64.5%、67.4%、73.3%。最後算出每個遊戲的平均分數做排行，以問卷的方式詢問本研究與Google Play之排行哪個較符合心目中的順序，有54.1%的人認為本研究較符合心目中的遊戲順序，且對於此推薦系統的適合程度有75%覺得適合，故透過蒐集巴哈姆特評價所做出的排行較遊戲下載平台的排行貼近使用者。

關鍵字

推薦系統 ； 文字探勘 ； 意見單元 ； 行動遊戲

並列摘要

Mobile games have already become one of the indispensable part of everyone’s life.Lots of people choose to or not to download the game basing on the comments other gamer gave it on Google Play or App Store . But for the purpose of better overall rank , some game producers use fake comments to cover the actual ones . While the 巴哈姆特 game forum provides a place for gamers to exchange their gaming experiences , not written by experts or game producers . The research selects some of the articles and information on the forum , finds out the articles that contains ratings and giving them points , finally sorts the games as a recommendation.
This research uses systems development as a methodology in information systems (is) proposed by Nunamaker(1991) and a few people , aiming to analyzing internet comments and sorting out the comments given by gamers. Finally we calculate the information and list the game rankings according to our calculation , making users understand more about the true ratings of the game , also , the users’ comments can be conveyed to the game producers as standards of improvement .
The objects of our study included 3 games: 七騎士,列王的紛爭,神魔之塔 which were all the top 6 games on Google Play revenue ranking list . The research information was collected from the articles on 巴哈姆特, and the articles were mainly in the sub-boards called “all articles” and “easy talking” in the forum. We found out the articles that contained ratings and gave them points ,also , we verified them manually. The accuracy rate were 64.5%, 67.4%, 73.3% . Then we calculated and came out with a game ranking according to their average points . We made survey questionnaires to find out which ranking bore more resemblance to their own ranking. The results showed that 54.1% of the respondents thought our research described better , also , the suitability of the recommending system reached 75% . As a conclusion , the comments and ratings collected from 巴哈姆特 are more corresponding than those collected from Google Play and App Store.

並列關鍵字

Recommendation Systems ； Text Mining ； Comments unit ； Mobile Games

參考文獻

[4] 張雅喬. (2015). 利用使用者評論幫助協同過濾的評分預測. 臺灣大學資訊工程學研究所學位論文.

[6] 李孟潔. (2009). 利用機器學習作法之中文意見分析.

[7] 林祐任. (2015). 未來性資訊檢索系統基於網路論壇之研究. 淡江大學資訊管理學系碩士班學位論文.

[9] 簡之文. (2012). 部落格文章情感分析之研究. 淡江大學資訊管理學系碩士班學位論文.

[13] 黃心宜. (2014). 基於影響力分析之意見單元評價的研究. 淡江大學資訊管理學系碩士班學位論文.