在企業之中，計算員工的績效、專案管理乃至營運管理的成本都是一件不容易的事情，通常需透過員工自行填寫工作日誌，但每位員工在填寫自己一天的工作內容時，可能會忘記一些零碎的工作，導致工作日誌內容遺漏，而專案的成本則需透過花費的人力、物力時間、交通、溝通等工作細項來計算，這樣的成本計算將會直接影響到公司的營運及決策，因此，企業需要針對營運成本做出精準的評估。
近年來的人工智慧相關技術的成熟，本論文基於人工智慧之技術，擬發展智慧化工作日誌系統的設計與實作。本論文透過員工線上會議的資料，對線上開會聊天的內容，利用LSTM文章分類與TF-IDF統計技術加以分析，以確定聊天內容所屬的專案，再透過發言人與時間的比對，協助員工自動化填寫每日工作日誌，且透過提供精準的工作日誌，還能計算專案中開會討論所花費之人事成本，進而以自動化及智慧化的方式，估算專案成本，提升企業管理專案成本與員工績效的準確性，讓企業控管營運成本時更有效率。

Today, many companies have changed the way of meetings from face-to-face to online, such as project meetings in the Line community software. In order to control the progress of the project and have better time and performance managements, the company usually requires each employee to write a daily work log to verify the performance of the work. However, employees need to recall the work content when writing the work log, which may result in incomplete work log due to easy forgetting. In addition, the cost of project is not easy to be predicted. Based on the above motivations, this thesis aims to solve the following challenges encountered in online meetings:
(1)	During the Line online meeting, employees may discuss many projects and cannot measure the time spent on each project accurately. This may cause the problem that the work log cannot reflect the length of work, and the cost of the project cannot be measured accurately.
(2)	Manually writing a work log is not complete and affects the assessment of the performance of the work.
(3)	It is difficult for managers to grasp the length of work and progress of employees' meetings, especially online meetings. The performance of each employee is not easy to evaluate.
(4)	The cost of project is difficult to be predicted.
In order to solve the above mentioned problems, this thesis a " Intelligent Assistant for Filling Working Diary and Project Cost Evaluation" approach, which applies Long Short Term Memory Network (LSTM) and TF/IDF techniques to take the text content of each chat room as its inputs during the Line meeting, identify which project the employee belongs to, and measure the time spent by the employee in the project, so as to assist the employee to write the work log and evaluate the cost of the entire project. The mechanisms proposed in this thesis expects to help better manage employee performance and reduce the workload of employee to fill up the daily work log as well as predict the project cost.

目錄
目錄	V
圖目錄	VI
表目錄	VII
第一章、簡介	1
第二章、相關研究	4
第三章、背景知識	7
3.1 LSTM技術	7
3.2 TF-IDF技術	9
3.3 餘弦相似性	10
第四章、系統架構	11
4.1 環境與問題描述	11
4.1.1欲解決的問題	11
4.1.2目標	11
4.2 系統架構	13
第五章、系統實作	26
第六章、實驗分析	29
第七章、結論	35
參考文獻	36
附錄-英文論文	37

圖目錄
圖 1：遞歸神經網路概念圖	8
圖 2：一般的遞歸神經網路與長短期記憶神經網路的比較圖	9
圖 3：設計及實作基於人工智慧之智慧化工作日誌系統	12
圖 4：系統架構圖	13
圖 5：訓練階段架構圖	14
圖 6：專案名稱與專案簡介資料示意圖	15
圖 7：LSTM模型訓練示意圖	16
圖 8：LSTM網路架構	17
圖 9：專案名稱計算階段	18
圖 10：LSTM專案辨識模組	19
圖 11：算各文檔TF-IDF與取關鍵字示意圖	20
圖 12：利用TF-IDF分類文檔的方法一示意圖	21
圖 13：利用TF-IDF分類文檔的方法二示意圖	22
圖 14：任務4的輸入、輸出與對應之模組	23
圖 15：專案工作時間計算階段	24
圖 16：輸入欲辨識的聊天室文檔	26
圖 17：LSTM辨識該文檔屬於各專案的機率	27
圖 18：TF-IDF關鍵字辨識模組計算的兩組機率	27
圖 19：專案名稱計算模組算出的機率	27
圖 20：工作時間計算模組統計出各員工花費於該專案的時間	28
圖 21：使用LSTM模型預測3個文檔屬於各個專案的機率	29
圖 22：關鍵字計算專案機率模組在不同關鍵字個數下的正確率	30
圖 23：餘弦相似度專案機率模組在不同關鍵字個數下的正確率	30
圖 24：LSTM模型與關鍵字計算專案機率模組透過變動α值與關鍵字個數產生的正確率	31
圖 25：LSTM模型與餘弦相似度專案機率模組透過變動α值與關鍵字個數產生的正確率	32
圖 26：關鍵字計算專案機率與餘弦相似度專案機率透過變動α值與關鍵字個數產生的正確率	33
圖 27：不同權重下系統的正確率	34

表目錄
表 1：相關研究功能比較表	6

參考文獻
[1]	Google,https://support.google.com/calendar/answer/37082?hl=zhHant&ref_topi-c=3417969 
[2] 	Tianxing He, Jasha Droppo "Exploiting LSTM Structure in Deep Neural Networks for Speech Recognition," 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Shanghai, 2016, pp. 5445-5449.
[3]	A. A. Hakim, A. Erwin, K. I. Eng, M. Galinium and W. Muliady, "Automated document classification for news article in Bahasa Indonesia based on term frequency inverse document frequency (TF-IDF) approach," 2014 6th International Conference on Information Technology and Electrical Engineering (ICITEE), Yogyakarta, 2014, pp. 1-4.
[4]	M. Abe, A. Hirayama and S. Hara, "Extracting daily patterns of human activity using non-negative matrix factorization," 2015 IEEE International Conference on Consumer Electronics (ICCE), Las Vegas, NV, 2015, pp. 36-39.
[5]	K. Hamdan, H. El Khatib and K. Shuaib, "Practical software project total cost estimation methods," 2010 International Conference on Multimedia Computing and Information Technology (MCIT), Sharjah, 2010, pp. 5-8.
[6]  F. A. Gers, J. Schmidhuber and F. Cummins, "Learning to forget: continual prediction with LSTM," 1999 Ninth International Conference on Artificial Neural Networks ICANN 99. (Conf. Publ. No. 470), Edinburgh, UK, 1999, pp. 850-855 vol.2. 
[7] 	Robertson, S. (2004), "Understanding inverse document frequency: on theoretical arguments for IDF", Journal of Documentation, Vol. 60 No. 5, pp. 503-520. 
[8] 	Wikipedia,“Cosine similarity”, https://en.wikipedia.org/wiki/Cosine_similarity#cite_-note-2