近年來網路社群非常活躍,非常多的網民都以社群媒體來分享與討論時事。不傴於此,網路上的群聚力量已經漸漸從虛擬走向現實,社群媒體的傳播力已經可以與大眾傳媒比擬。像台大 PTT 的八卦版就是一個這樣具指標性的社群媒體,許多新聞或是事件都從此版開始討論,然後擴散至主流媒體。透過觀察, 網路鄉民常常會以略帶灰諧的方式,發明新的詞彙去討論時事與人物,例如: 割闌尾、祭止兀、婉君、貫老闆...等。這些新詞的出現,很可能代表一個新的熱門話題的正在醞釀中。但若以傳統的關鍵詞搜索,未必能找到這些含有此類新詞的討論文章。因此,本研究提出一個基於「滑動視窗(Sliding window)」的技巧來輔助中文斷詞,以利找出這些新詞,並進而透過這些新詞對來探詢社群媒體中的新興話題。我們以此技巧修改知名的Jieba 斷詞工具,加上新詞偵測的機制,並以 PTT的八卦版為監測對象,經過長期的的監測後,結果顯示我們的系統可以正確的找出絕大多數的新詞。此外,經過與主流媒體交叉比對,本系統發現的新詞與新話題的確有極高的相關性。

In recent years, a very large portion of Internet users are used to share and discuss current events over social media. Indeed, as more and more people actively participate in the various virtual communities over the Internet, it is fair to say that the spread power of social media can be compared with that of mass media. The popular PTT gossip board is one such indicator of social media. In Taiwan, many news or events originated in this board would spread to the mainstream media and then become hot topics in the society.
We notice that many Internet users often invent new vocabulary to discuss current events and characters. The emergence of these new words may later grow into a new hot topic in the society. However, if we apply the traditional keyword search, we may not be able to find these articles with such new words. Therefore, this thesis present a "sliding window" technique to assist Chinese segmentation tool for facilitating the identification of these new words. Besides, these new words often represent a key indicator for new discussion topics. We use this technique to extend the famous Chinese segmentation tool, Jieba, with a new word detection mechanism, and apply it to the PTT gossip board. After a long-term monitoring, we obtain the results showing that our system can correctly identify the vast amount of new words in the board. In addition, after a cross comparison with the mainstream media, the new words identified by this system are indeed related to the popular social topics in a very high manner.

[1] Chen. & Bai. (1998), Unknown word Detection for Chinese by Corpus-based Learning Method.
[2] Chen. & Ma. (2002), Unknown Word Extraction for Chinese Document.
[3] L. Jin. (2013), Number in Chinese: A Corpus-Based Computational Investigation.
[4] QX Lin. (2010), 結合長詞優先與序列標記之中文斷詞研究。
[5] Yi-Lun Wu. (2011), 多語語碼轉換之未知詞擷取。
[6] Zhihui. Wu, Hongwei. Liu, Li. Chen. (2014), 统计学与应用, 3, 30-35, 高效朴 素贝叶斯 Web 新闻文本分类模型的简易实现,The Simply Implement of Effective Statistical and Application
[7] Z. Wu (2014), The Simply Implement of Effective Naive Bayes Web News Text Classification Model
[8] 陳鍾誠、許聞廉. (1998), 結合統計與規則的多層次中文斷詞系統
[9] 陳聰宜. (2012), 新聞事件偵測與追蹤結合時間區間之分群分類演算法評比
[10] DAG, http://www.csie.ntnu.edu.tw/~u91029/DirectedAcyclicGraph.html#1
[11] ACID, http://zh.wikipedia.org/wiki/ACID
[12] MongoDB, http://docs.mongodb.org/manual/core/crud-introduction/
[13] NoSQL, http://zh.wikipedia.org/wiki/NoSQL
[14] Jieba 斷詞工具, https://github.com/fxsjy/jieba
[15] 隱馬可夫模型, http://zh.wikipedia.org/wiki/隐马尔可夫模型
[16] 维特比算法, http://zh.wikipedia.org/wiki/维特比算法
[17] 洪仲丘案在 PTT, http://zh.PTTpedia.wikia.com/wiki/%E6%B4%AA%E4%BB%B2%E4%B8%9 8%E4%BA%8B%E4%BB%B6%E5%9C%A8PTT