基於人工智慧之廣告關鍵字與文案自動生成技術

Automatic Generation Technology of Advertising Keywords and Copywriting based on Artificial Intelligence

李旻祐(Min-Yu Lee)

指導教授 ： 張志勇

共同指導教授 ： 郭經華(Chin-hwa Kuo)

淡江大學/工學院/資訊工程學系碩士班/碩士(2020年)

https://doi.org/10.6846/TKU.2020.00162

摘要

在網路上投放一則廣告，只需要準備商品網站內容、關鍵字、標題、說明文案以及價錢，就能夠將自己的廣告，投放於很多用戶都會造訪的網站，如 Google或Facebook等網站及平台，讓自己的商品，能達到能夠得到更多的宣傳。然而，針對一個廣告內容而言，要找出較佳的關鍵、標題及文案說明，並不容易，以往在搜尋引擎投放廣告，必需藉由專業人士，設計廣告文案，才能以較有效的方式來投放廣告。
隨著現今網路廣告平台越來越多如，Google、YouTube及Facebook等大型平台都提供了網路廣告的服務，使得想要投放廣告的商家日益增多，讓廣告公司人力不足，但要培養一個廣告專家是不容易的件事情，需要長時間的訓練以及經驗累積才能與其他的廣告公司競爭。
因此本研究擬透過AI技術，扮演文案專，或協助文案新手，對擬投放的廣告，自動找出關鍵字、文案標題及文案說明。本研究關鍵字提取方法綜合了三種元素，第一種，網頁內的文字，透過TF-IDF提取本文的關鍵字，這樣能夠符合廣告關鍵字需要與內文相關的需求；第二元素，則是專家的經驗，透過爬蟲抓取Google 搜尋廣告的前三名廣告內容字詞做為關鍵字；第三則是使用者常用的熱門搜尋字詞，我們透過Google Trend熱門趨勢，抓取與網頁內文有關的熱門字詞，藉此提高廣告被搜尋到的次數，我們將這三種元素綜合起來，做為提取廣告關鍵字依據。除了關鍵字提取外，本研究自動生成標題與說明的技術，透過兩種方式，第一種為生成式的，一樣包含了，專家與熱門的元素，透過專家提供的過往廣告說明與標題，與Google搜尋前三名的廣告，透過實體辨識分類句子，並透過替換字詞的方式，將過往的標題與說明，經過演算法選出與網頁相關的標題，並根據前後字詞關係進行字詞替換。
本研究與以往的關鍵字提取與說明標題生成不同，我們的關鍵字提取方法，參考了外來的元素，藉此提高關鍵字的多樣性，並且能自動化產生。實驗結果表明，本論文所提出的方法可以成功地提供滿足精度和覆蓋率的關鍵字。

關鍵字

關鍵字建議 ； 關鍵字提取 ； 標題生成 ； 說明生成 ； 廣告應用

並列摘要

To placement an advertisement on the Internet, you only need to prepare the content of the product website, keywords, title, description and price, and then you can place your own advertisement to get more publicity for your product. However, it is not easy to find the best keywords, titles, and descriptions for an advertisement content. In the past, search engine advertising required professionals to design ad copy in order to place ads in a more effective way. Nowadays, there are more and more online advertising platforms such as Google, YouTube and Facebook. This has led to an increasing number of businesses wanting to place advertisements, leaving advertising agencies short of manpower. But it is not easy to develop an advertising expert, it takes long time training and experience to compete with other advertising agencies.
This study intends to use AI technology to act as a copywriter, or to assist novice copywriters to automatically find keywords, titles, and descriptions of the advertisements they intend to place. The keyword extraction method of this study integrates three elements: the first one is the text in the webpage, and the keywords of this article are extracted through TF-IDF, which can meet the needs of keywords and text related to advertisements; the second element is the experience of experts, who extract the top three ad content words of Google search ad through crawlers as keywords; the third element is the popular search terms used by users. In order to increase the number of ad searches, we use Google Trend hot trends to capture popular words related to webpage content. In addition to keyword extraction, this study automatically generates headlines and descriptions in two ways, the first one is generative, which contains both expert and popular elements, through the past ad descriptions and headlines provided by experts, and Google search the top three ads, through entity recognition classification sentences, and through word substitution. The headings of the web pages are related to the keywords, and the word substitution is performed according to the relationship between the first and last words. This study differs from previous keyword extraction and descriptive headline generation. Our keyword extraction method makes reference to external elements in order to increase the diversity of keywords and automate the generation.
This study differs from previous keyword extraction and descriptive title generation in that our keyword extraction method, which makes reference to exogenous elements in order to enhance keyword diversity, can be automated and generated. The experimental results show that the proposed method is successful in providing keywords that satisfy both accuracy and coverage.

並列關鍵字

Keywords suggestion ； Keywords extraction ； Search advertising ； Advertisement keyword suggestion and copywriting application

參考文獻

[1]P. Sun, L. Wang and Q. Xia, "The Keyword Extraction of Chinese Medical Web Page Based on WF-TF-IDF Algorithm," in 2017 International Conference on Cyber-Enabled Distributed Computing and Knowledge Discovery (CyberC), Nanjing, Oct,2017, pp. 193-198.

[2]A. Joshi and R. Motwani, "Keyword Generation for Search Engine Advertising," in Sixth IEEE International Conference on Data Mining - Workshops (ICDMW'06), Hong Kong, Dec, 2006, pp. 490-496.

[3]H. Nie, Y. Yang and D. Zeng, "Keyword Generation for Sponsored Search Advertising: Balancing Coverage and Relevance," in IEEE Intelligent Systems, vol. 34, no. 5, pp. 14-24, Sept. 2019.

[4]Y. Yang, B. J. Jansen, Y. Yang, X. Guo and D. Zeng, "Keyword Optimization in Sponsored Search Advertising: A Multilevel Computational Framework," in IEEE Intelligent Systems, vol. 34, no. 1, pp. 32-42, 1 Jan ,2019.

[5]Himat Shah, Mohammad Rezaei, and Pasi Fränti. "DOM-based keyword extraction from web pages," in Proceedings of the International Conference on Artificial Intelligence, Information Processing and Cloud Computing (AIIPCC '19). Association for Computing Machinery, New York, USA, Dec, 2019,no 62, pp. 1–6.