網路情感分析對於手機應用程式評價之影響的研究

A Study of the Impact of the Internet Sentiment Analysis on APP Rating

酆偉寬(Wei-Kuan Fong)

指導教授 ： 蕭瑞祥

淡江大學/商管學院/資訊管理學系碩士班/碩士(2015年)

https://doi.org/10.6846/TKU.2015.00674

摘要

根據Surikate 與GfK調查 (2013)顯示，85%使用者會在下載應用程式(Application, APP)前，參考在APP STORE(iOS Application Store, APP STORE)上其它用戶對該APP的評價，這些參考項目包含該APP的預覽圖、銷售價格與內文評論等，可知使用者通常會參考其它用戶的使用經驗以決定是否下載該應用程式。
根據資策會FIND調查我國智慧型行動裝置持有人數於2014年底已達1400萬以上，又依據LINE的官方統計，在我國LINE的下載次數於2014年6月中突破1700萬大關，下載次數冠居所有APP；從數據上來看，LINE在我國行動族群間之普及率早已超過九成，但該軟體在APP STORE中的用戶評分卻僅有2.4分，足見用戶評分與下載次數並無決定性的相互影響關係，且難以反應出真實的評價水準。
據此，本研究欲以探討使用者對於手機應用程式的真實評價為研究目的，期望能夠透過蒐集網路評論文本，將評論文本轉化為特徵量化數據，輔以結合人工判讀內容並加以分析後，提出一可反應APP真實評價之評分模型。
為了驗證本研究提出之評分模型效能，本研究亦將透過系統發展研究法，以支持向量機(Support Vector Machine, SVM)分類模型將網路評論文本依特徵項轉化為特徵量文本後，進行訓練與測試，再將支持向量機分類模型產出之分類結果比對人工判讀之結果，計算其準確率做為系統驗證的依據。
經由實驗結果顯示，透過將詞彙做為特徵值，評分模型的整體分類準確率將近九成，顯示本研究之APP評分模型有良好的反應能力；期望本研究之研究結果能在後續相關領域下之研究有所貢獻。

關鍵字

意見探勘 ； 情感分析 ； SVM

並列摘要

According to Surikate and GfK (2013), 85 percent Application (APP) users refer to the comments and ratings of other users on iOS Application Store, APP STORE before actually downloading the APP. It can be noticed that people commonly examine the user experience of others when determining whether to download the product.
The investigations conducted by the FIND (2014) states that people who possess a smartphone in our nation has risen over 14 million. The statistics from the LINE official website (June, 2014) also indicates that its downloads had made a significant breakthrough of over 17 million, making LINE the most downloaded APP. From this information, we can conclude that LINE has a penetration rate of over 90 percent among Taiwanese cellphone users. However, this APP is rated a 2.4 in the APP STORE, implying that user ratings does not determine the amount of downloads. Moreover, this score limitedly reflects the real value of the product.
Based from this observation, the objective of this research is to discuss the authentic ratings of cellphone APP users. Via adapting a scientific method, the collection of APP word ratings on the internet will be converted into quantized data. Analysis will be done with the assistance of manual interpretations in order to propose a modal that adequately reflects the user experience.
For the purpose of verifying the efficiency of this rating model, this study is built on the System Development Methodology. After the wording of internet comments are transformed into characterized quantized data, it is then entered into the classification of the Support Vector Machine (SVM) for further training and testing. The data that is classified by the SVM is compared with the outcome of manual interpretations so that to calculate the accuracy rate.
The results of this experiment indicate that through corpus categorizing, this modal reached an accuracy rate of 90 percent. This demonstrates that this APP rating modal has outstanding reflection ability. Thereby, may the results of this research contribute to future studies in this field.

並列關鍵字

Opinion mining ； Sentiment analysis ； SVM

參考文獻

[5] 黃婷穎、莊裕澤，《網路評價視覺化》，碩士論文，國立台灣大學資訊管理研究所，2011。

[9] 簡之文、蕭瑞祥，《部落格文章情感分析之研究》，碩士論文，淡江大學資訊管理研究所，2012。

[11] 謝衫蒂、戴敏育，《應用機器學習與多辭典的中英雙語意見分析之研究》，碩士論文，淡江大學資訊管理研究所，2014。

[1] 楊盛帆、陸承志，《以整合式規則來做網路論壇上的3C 產品口碑分析》，碩士論文，元智大學資訊管理研究所，2009。

[3] 邱鴻達、梁婷，《意見探勘在中文電影評論之應用》，碩士論文，國立交通大學資訊工程研究所，2011。