| dc.description.abstract | 隨著科技的日新月異，我們隨時都要精進自己以獲取新知，避免被世界淘汰，於是帶動諸如Stack Overflow, Yahoo Answers, Quora, Zhihu (知乎)等社群問答網站(Community Question Answering，CQA)的興起。使用者可以在上面提問、回答問題，作為彼此交流與學習的平台。
雖然社群問答網站的興起帶給使用者很大的便利，但是由於問題數量眾多，多數問題通常杳無音訊，想要及時得到問題正確的回覆，不可否認需要運氣與時間的等待。我們認為，若可於CQA 網站中正確地找出專家，則可藉由把對的問題推薦給有能力回答的專家，便可提升使用者互動，解決問題之效率。
本研究首先透過非監督方法 -- Yang, Liu, et al. (2013)所建的TEM (Topic Expertise Model) 模型，擷取使用者對每個主題下專精程度的特徵向量，並利用History post embedding，以詞嵌入(Word Embedding)的特性，擷取語意程度的特徵向量，再利用問題與回答者之相似度作為推薦專家之依據。我們鎖定Stack Overflow (世界前幾大的程式設計領域的問答網站)作為研究目標，並獲得良好之準確率，並期望研究成果可於其他CQA 網站使用。
本篇論文的貢獻是將TEM模型與詞嵌入的歷史資訊做結合,當在社群網路結構並非那麼完整時有效的把對的問題配對給對這個問題有能力回答的專家以提升社群網路參予度低的問題。
| zh_TW |
| dc.description.abstract | With the ever-changing technology, we humans have to be willing to keep on learning in order to avoid being demoted by the world. Therefore, the reasons above led to the rise of the community question answering websites, such as Stack Overflow, Yahoo Answers, Quora, Zhihu (知乎), and so on and so forth. Users can ask questions, answer questions, exchange and discuss ideas with each other in the above platform.
Although the rise of community question answering websites can surely bring great convenience to users, there is still room for improvement. Due to the large numbers of questions, most questions usually receive no response or get inappropriate answers. It is without doubt to rely on luck and time to get correct answers in time. Therefore, we believe that if we can find experts precisely in CQA websites, we can improve the efficiency of the participation rate by routing right questions to experts.
In this study, we firstly utilize TEM (Topic Expertise Model), which is an unsupervised model published by Yang, Liu, et al. (2013), for capturing the degree of expertise of question and answerer under different topic. Furthermore, we utilize History Post Embedding, which is published in this thesis by using word embedding techniques, to extract semantic meanings to enhance the understanding of question sets. Finally, we combine the vector of topical expertise with History Post Embedding and perform a recommendation formula to rank experts. We target Stack Overflow, which is one of the biggest computer programming field CQA websites in the world, as our research goal and obtain good result. Moreover, we expect the research result to be available on other CQA websites.
The main contribution of this thesis is combining TEM model with distributed representation of user historical information which can solve the problem of low participation rate in CQA websites when social network structure is not so complete.
| en_US |