請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/83251
標題: | 基於BERTSUM的搜尋引擎最佳化商品標題生成模型 Search Engine Optimization For Product Title Generation Model Based On BERTSUM |
其他標題: | Search Engine Optimization For Product Title Generation Model Based On BERTSUM |
作者: | 黃翔岳 Hsiang-Yueh Huang |
指導教授: | 莊裕澤 Yuh-Jzer Joung |
關鍵字: | 文本摘要生成,預訓練模型,深度學習,電子商務,搜尋引擎最佳化, Text Summarization,Pre-trained Model,Deep Learning,E-commerce,Search Engine Optimization, |
出版年 : | 2022 |
學位: | 碩士 |
摘要: | 隨著網際網路的發展,各類產品的銷售管道已不侷限於實體店面販賣,透過電子商務,商家能接觸到比實體店面更多的顧客。隨著PCHOME、淘寶等電子商務平臺的出現,企業或一般民眾都能在平臺上販賣產品。由於同質性產品間競爭激烈,商家為了提升產品的曝光度進而增加電商轉換率,常會針對電子商務平臺設計搜尋引擎最佳化的標題,此類標題通常會以簡潔有力的方式表達產品的特色並吸引消費者目光,以及增加產品被搜索機會的字詞,其目的皆為使產品能在搜尋結果頁中占據較好的排名。但電子商務產品種類繁多,以人力撰寫搜尋引擎最佳化標題較為繁雜,且不同種類的產品標題有其偏重關注的特色。為解決上述問題,本研究將標題生成任務類比為文字摘要生成任務,使用深度學習技術實作一個搜尋引擎最佳化標題生成系統,使用者輸入產品敘述文案後,系統即可生成適合該產品的搜尋引擎最佳化標題。本研究使用BERTSUM預訓練模型,並以TaoDescribe商品敘述資料集訓練系統自產品敘述文案生成搜尋引擎最佳化標題的能力。而在最後的實驗結果中,本系統在自動評估上與其他應用於不同任務的模型有著相當的表現。在搜尋引擎最佳化方面,本系統的生成標題在搜尋結果頁的排名上與原始標題表現相當,且針對不同種類的產品皆可生成符合該產品類別特性的標題。 With the development of the Internet, the sales pipeline of various products is no longer limited to physical stores. Through e-commerce, merchants can reach more customers than physical stores. With the emergence of e-commerce platforms such as PCHOME and Taobao, enterprises or ordinary people can sell products on the platform. Due to the fierce competition among homogeneous products, e-commerce merchants often use search engine optimization (SEO) techniques to improve product titles to enhance product exposure and increase conversion rates. These titles are usually simple and powerful in expressing the product's features and attracting consumers' attention. They also tend to use words that can help product to be searched by search engines. However, there are many different types of e-commerce products, and writing SEO titles is quite a challenge to human. Besides, different types of product titles have different characteristics that favor search engine attention. In order to solve this problem, our study analogizes title generation task to text summary generation task, and uses deep learning technology to implement a SEO title generation system. We use BERTSUM pre-training model and TaoDescribe product description dataset training system to generate SEO titles from product description. The results show that our system performs comparably to other models applied to different tasks in terms of automatic evaluation. In terms of search engine optimization, our system generates headlines that are ranked by search engines similarly to the ranks of the original titles. The system also performs well for different types of products. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/83251 |
DOI: | 10.6342/NTU202300134 |
全文授權: | 同意授權(全球公開) |
顯示於系所單位: | 資訊管理學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
U0001-0518230116585013.pdf | 1.58 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。