中文短回文立場分類

Abstract

隨著社交網站的崛起與普及，資訊分享和傳播媒介已經大幅地改變，每一個個體在網路上傳播的資訊或發表的言論，都可能是造成某個運動成功、某個事件受到矚目、或者某件作品、影片在網路上大紅大紫的部分原因。我們在網路上的一言一行，都能夠影響到其他人的看法，也留下可供觀察的足跡，這些網路上的資料及其影響力，開啟了一連串關於網路行為的研究。站在語言學的角度，我們決定觀察使用者在網路上如何表達立場、以及如何說服他人。 本研究旨在網路上的立場表達行為，以批踢踢的回文作為觀察語料，建立使用上能夠透露發言者對於文章或是命題所採態度的語言線索。研究過程包含從批踢踢的推文和噓文當中，找出主觀性用語以及論辯用語，再以這些用語輔助文本自動分類，將自動立場分類的表現與基線(baseline)作比較，實驗結果顯示本研究所建立的語言線索能夠幫助分類器的表現提升，最高能達到百分之二十，後續研究若能持續擴大標記的語料筆數，並且能夠掌握語境與語言線索之間的關係，將能夠更大幅度地提升分類準確性。

With the development of social networking services, information sharing and the form of media have been revolutionized. Each of us might be part of the reason why certain movements are able to take place, why several issues are finally noticed by the public, and how a video or a piece of work goes viral. What we say or do on the Internet has its influence on others, and leave traces that we can observe. Thus gave rise to studies that aim at understanding online behaviors. From a linguistic point of view, we find it worthwhile to observe how people take a stance online and their attempts to persuade others. The current study aims at observing stance-taking behavior on short comments on PTT in order to establish resources on “linguistic cues” that reveal a speaker’s overall position on an article or on a proposition. Subjective cues and arguing cues are identified from extracted PTT positive and negative comments. The cues are then used to assist in automated stance classification to compare with baseline performance. Results indicate that the cues can help raise up to 20 percent of accuracy. The classifier can be better improved with larger tagging set and techniques that can identify context in future work.