基於檢索與閱讀理解的聊天機器人之研究與實作 - 以稅務諮詢為例

The Study and Implementation of Chatbot Based on Retrieval and Reading Comprehension – Take Tax Consulting as An Example

簡琬真(Wan-Chen Chien)

指導教授 ： 孫春在

國立交通大學/資訊學院/資訊學院資訊學程/碩士(2020年)

摘要

台灣的中小企業佔全國工商業的總家數九成以上，這些規模較小的企業通常都將會計帳務處理、稅務結算申報、稅務諮詢等業務委託給代客記帳的會計事務所，不論是中小企業負責人或一般民眾對於稅務上的任何問題，除了可以搜尋政府財稅機關官網上的公開資訊外，便是透過專業會計人員來獲取知識。
　　民眾如果對稅務有疑問，一般都需要特別向稅務專業諮詢人員申請諮詢時間，而稅務專業諮詢人員往往沒有精通各種稅法，再加上稅務法規修法後，資料未即時更新，民眾自行查閱資料時，不容易分辨資料的時效性，因此，本研究的最終目地是開發一套深諳國稅、地方稅所有稅法的稅務諮詢聊天機器人，除了能有效率地為民眾解惑常見問題以外，也能為稅務機關帶來新服務型態的轉變。
　　本研究運用爬蟲技術將各大財稅機關網站、財稅新聞網站上的資料存入聊天機器人的知識庫，包含有稅務問題、稅務法規、稅務新聞等各類文件。本論文運用資訊檢索及深度學習技術萃取出問題的答案，整合各模型演算法的優點，建構並實作出稅務諮詢問答系統，使聊天機器人能提升稅務諮詢的服務品質。

關鍵字

聊天機器人 ； 深度學習 ； 人工智慧 ； 自然語言處理

並列摘要

More than 90% of industrial and commercial households are Small and Medium Enterprises. These smaller companies usually entrust business such as accounting, settlement declaration, tax consulting to a bookkeeping accounting firm.
　　For any tax issue, business owner or general public search for public information on some official website of financial and tax authorities and gain knowledge through professional accountants. People have any questions about taxes need to schedule an appointment for consultation. But tax professional consultants are often not proficient in various tax laws. In addition, after the amendment of tax laws and regulations, the information is not updated in real time, and it is not easy to distinguish the validity of the data when the people ask for the information by themselves.
　　Therefore, the goal of this paper is to develop a tax consulting chatbot who are proficient in various tax laws, including national and local taxes. In addition to being able to effectively solve common problems for the people, it can also bring new service type changes to tax authorities.
　　This work will crawl some pages from some financial and news related official sites to store them into a tax knowledge-based corpus, including some document types of tax issues, tax regulations, and tax news. This paper refers to some algorithms used in the well-known reading comprehension dataset for many related applications. Context would be training through natural language processing to enhance semantic analysis. To integrate the advantages of each model algorithm of information retrieval technology and deep learning model to implement a tax consulting question answering system to enable chatbot to improve the quality of tax advisory services.

並列關鍵字

chatbot ； Deep Learning ； AI ； NLP

參考文獻

doi:10.1093/comjnl/1.1.36, 1950s.

[42] Sparck Jones, K." A Statistical Interpretation of Term Specificity and Its Application in Retrieval.". Journal of Documentation. 28: 11–21. doi:10.1108/eb026526. 1972.

[52] Breitinger, Corinna; Gipp, Bela; Langer, Stefan. " Research-paper recommender systems: a literature survey". International Journal on Digital Libraries. 17 (4): 305–338. doi:10.1007/s00799-015-0156-0. ISSN 1432-5012. 26 July, 2015.

[1] 經濟部中小企業處

https://www.moeasmea.gov.tw/article-tw-2276-4766，2019/11/05