以留言內容為基礎之留言主題趨向分析模式—以文教事件新聞為例

Trend Analysis for Comment Topics -A Case Study of Comments related to Educational News

鄭郁彬(Yu-Pin. Cheng)

指導教授 ： 侯建良

國立清華大學/工學院/工業工程與工程管理學系/碩士(2017年)

摘要

當學校行政人員欲了解某校務事件新聞的輿論效應，其往往透過網際網路搜尋該事件新聞之留言，再瀏覽全部留言後主觀地運用相似留言中之特徵判斷留言的主題，之後再仔細瀏覽各留言並主觀地判斷各留言所隸屬之主題，以取得該校務事件新聞之輿論趨勢，進而協助學校高層官員構思與作出具體決策。然而，現今新聞平台或社群平台均無法自動分析某校務事件新聞所引起之留言的留言主題，以致學校行政人員往往需耗費大量時間閱讀與理解全部留言內容並主觀訂定留言主題及其特徵，再重新閱讀各留言內容以主觀判斷各留言之特徵與各留言主題之特徵的相似度，以將各留言歸屬至對應留言主題，進而了解該校務事件新聞之輿論傾向。
為解決學校行政人員了解留言輿論趨向所面臨之問題，本研究乃先透過前置階段蒐集文教事件新聞所衍生之留言，並整理留言之特徵值、留言主題判斷原則與留言所隸屬之留言主題判斷原則。之後，本研究根據置階段之解析結果，本研究發展一套「留言主題趨向分析」方法論，而此方法論主要包含「留言內容特質擷取」、「留言主題判定」、「留言所隸屬之留言主題判定」與「留言趨向分析」等四大階段。其中，「留言內容特質擷取」階段可將網際網路所蒐集之留言依留言內容擷取特徵點（即可辨識其特質之詞彙）；之後，「留言主題判斷」階段乃依留言內容特質擷取結果比較各留言間之關聯性，以判別留言主題並建立各留言主題所對應之特質；接著，「留言所隸屬之留言主題判斷」階段乃透過比較各留言主題的特質與各留言內容的特質擷取結果之相似度判斷各留言所應隸屬之留言主題；最後，「留言趨向分析」階段則將各留言主題所對應之留言數以視覺化方式呈現，以利讀者快速且準確地判斷所關注之新聞事件所對應之目前輿論趨向。
未來，若學校行政人員欲了解某校務事件新聞之輿論傾向時，即可透過此些事件新聞之留言所對應的留言主題趨向視覺化圖式結果，釐清該校務事件新聞所對應之主題趨向，進而協助學校高層長官快速且準確構思出針對該校務事件新聞的具體對策。

關鍵字

輿論趨向分析 ； 留言特徵分析 ； 視覺化分析方法

並列摘要

To know the trend of a specific news related to school, staff firstly tend to search for the comments related to the news, then read and analyze all the comments to find the characteristics of comments related to educational news. Secondly, they will review all the comments to shape the topics and categorized all the comments into the above topics. However, there are two problems exists currently. First of all, current social media platform cannot automatically analyze the trend of comments related to a specific school news. Second, staff need to waste a lot of time reading and juding the topics of comments related to a specific news so as to get the trend of a specific school news.
To solve the above problem, we proposed a preparation stage to figure out the characteristics of comments related to educational news, the principles of generating topics related to comments and the principles of judging the topics of comments. After that, we proposed a Trend Analysis of Comment Topics Methodology, it includes four stages, which are collecting comment characteristics, judging comment topics, judging the topic of each comment, analyzing comment trend. As for collecting comment characteristics stage, it can collect the six characteristics of each comment. As for judging comment topics stage, it can compare the relationship among all the comments. As for judging the topic of each comment stage, it can categorize each comment into its topic. As for analyzing comment trend stage, it can show the graph of comment topics. This methodology we proposed can assist user to get the trend of a specific news related to school fastly and accurately.
In the future, we hope the methodology and system we proposed can help staff can use the methodology and the related system to analyze the trend of a specific news, know the trend of a specifc news by the multiple graph related to the topics, make a quick decision in reponse to the specific news related to school.

並列關鍵字

comment trend analysis ； comment characteristic analysis ； graphical analysis

參考文獻

1. Akaichi, J., 2013, “Social networks’ Facebook’ statutes updates mining for sentiment classification,” Social Computing, Vol. 1, pp. 886-891.

2. Amit, C., Rahul, R. and Savita, A., 2013, “A new approach to detect and extract characts from off-line printed images and text”, Procedia Computer Science, Vol.17, pp. 434-440.

3. Amrutha, B. and Mintu, P., 2015, “Keyword based Tweet extraction and detection of related topics,” Procedia Computer Science, Vol. 46, pp. 364-371.

4. Barnaghi, P., Ghaffari, P. and Breslin, J. G., 2016, “Opinion mining and sentiment polarity on Twitter and correlation between events and sentiment,” IEEE Big Data Computing Service and Applications, pp. 52-57.

7. Cai, H. Yang, Y., Li, X. and Huang, Z., 2015, “What are popular: Exploring Twitter features for event detection, tracking and visualization,” Proceedings of the 23rd ACM international conference on Multimedia, pp. 89-98.