在目前產品數量眾多、電商發達的環境當中，如何吸引消費者目光是一件很重
要的事情，而客製化的產品描述就是其中一種方法，也因為商品眾多，人工所產生
的文案效率有限，所以自動客製化的產品描述文案生成，對現在電商發達的環境當
中很有潛力。以往的產品描述文案生成，大多都是用 Encoder-­Decoder 的模型架構，
但是這種基礎架構有兩個需要改進的問題缺點：(1) 生成的產品文案字詞容易重複性
高，相同類似產品給予不同人的描述文案都非常地近似與狹隘，達不到針對性與客
製化的效果。(2) 無法精確的描述出產品名稱品牌，用詞過於通用。本研究運用兩種
方法來讓上述問題得以解決，首先運用 Select­-Attention 機制來幫助模型學習並決定
客戶與產品的混合特徵，再藉此輸出客製化的產品描述文案。除此之外，本研究還
透過 Copy 機制，讓模型在產生產品文案時也會參考到產品名稱本身的訊息，得到更
精確的產品名稱描述。綜上所述，本研究的貢獻如下：(1) 使用 Select-­Attention 機制
讓模型可以學習如何產生產品與客戶屬性的混合特徵，藉此產生出客製化描述文案。
(2) 利用 Copy 機制達到更精準的產品名稱描述，並且在 BLEU 的指標上能有所提升。

關鍵字: 產品描述生成, 文章摘要生成, Select-Attention 機制, Copy 機制

In the current environment of electronic commerce with a large number of products, it is very important to attract the attention of consumers, customized product descriptions are one of the methods, and because of the large number of products, the efficiency of manual copywriting is limited, so the automatic customized product description copywriting generation has great potential in the current e-commerce environment. In the past, most of the product descriptions were generated using the Encoder-Decoder model framework. However, there are two shortcomings of this basic framework that need to be improved:(1) generated product descriptions are easily duplicated, and the product descriptions for the same product are very similar and narrow to different people, which can not achieve the effect of targeting and customization.(2) it is difficult to describe the product name and brand precisely, and the words used are too generic. This study uses two approaches to solve the above problem. First, we use the select-attention mechanism to help the model learn and determine the mixed features of customers and products, and then output customized product descriptions. Second, copy mechanism was used to allow the model to generate product descriptions with reference to the product's name, resulting in more accurate product descriptions. In summary, the contributions of this study are as follows:(1) Using the select-attention mechanism allows the model to learn how to generate a mixture features of products and customer's attribute to produce customized descriptions.(2) Use copy mechanism to achieve more accurate product descriptions and improve on BLEU metrics.

Keyword: product description generation, text summarization, select-attention
mechanism, copy mechanism

Aishwarya Agrawal, Jiasen Lu, Stanislaw Antol, Margaret Mitchell, C. Lawrence Zitnick, Dhruv Batra, and Devi Parikh. Vqa: Visual question answering. 2016.

Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to align and translate. 2016.

Peter F. Brown, John Cocke, Stephen A. Della Pietra, Vincent J. Della Pietra, Fredrick Jelinek, John D. Lafferty, Robert L. Mercer, and Paul S. Roossin. A statistical approach to machine translation. Computational Linguistics, 16(2):79–85, 1990.

Zhangming Chan, Xiuying Chen, Yongliang Wang, Juntao Li, Zhiqiang Zhang, Kun Gai, Dongyan Zhao, and Rui Yan. Stick to the facts: Learning towards a fidelityoriented E­commerce product description generation. pages 4959–4968, November 2019.

Zhangming Chan, Yuchi Zhang, Xiuying Chen, Shen Gao, Zhiqiang Zhang, Dongyan Zhao, and Rui Yan. Selection and generation: Learning towards multi-­product advertisement post generation. pages 3818–3829, November 2020.

Qibin Chen, Junyang Lin, Yichang Zhang, Hongxia Yang, Jingren Zhou, and Jie Tang. Towards knowledge-­based personalized product description generation in e­-commerce. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery Data Mining, Jul 2019.

Kyunghyun Cho, Bart van Merrienboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. Learning phrase representations using rnn encoder­-decoder for statistical machine translation. 2014.

Junyoung Chung, Caglar Gulcehre, KyungHyun Cho, and Yoshua Bengio. Empirical evaluation of gated recurrent neural networks on sequence modeling. 2014.

Antonio M. Corbí­Bellot, M. Forcada, Sergio Ortiz Rojas, Juan Antonio Pérez, Gema Ramírez-­Sánchez, F. Sánchez-­Martínez, I. Alegria, A. Mayor, K. Sarasola, and Euskal Herriko Unibertsitatea. An open­-source shallow­-transfer machine translation engine for the romance languages of spain. 2005.

Jeffrey L. Elman. Finding structure in time. Cognitive Science, 14(2):179–211, 1990.

Jianfeng Gao, Michel Galley, and Lihong Li. Neural approaches to conversational ai. 2019.

Karl Moritz Hermann, Tomáš Kočiský, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suleyman, and Phil Blunsom. Teaching machines to read and comprehend. 2015.

Sepp Hochreiter and Jürgen Schmidhuber. Long short-­term memory. Neural Comput., 9(8):1735–1780, November 1997.

Minh-­Thang Luong, Hieu Pham, and Christopher D. Manning. Effective approaches to attention-­based neural machine translation. 2015.

Junhua Mao, Wei Xu, Yi Yang, Jiang Wang, and Alan L. Yuille. Explain images with multimodal recurrent neural networks. 2014.

Michael L. Mauldin. Semantic rule based text generation. page 376–380, 1984.

Jianmo Ni and Julian McAuley. Personalized review generation by expanding phrases and attending on aspect­-aware representations. pages 706–711, July 2018.

Alexander M. Rush, Sumit Chopra, and Jason Weston. A neural attention model for abstractive sentence summarization. 2015.

Abigail See, Peter J. Liu, and Christopher D. Manning. Get to the point: Summarization with pointer­-generator networks. 2017.

E. Sumita and H. Hda. Example-­based nlp techniques - ­a case study of machine translation -­. 1992.

Ilya Sutskever, Oriol Vinyals, and Quoc V. Le. Sequence to sequence learning with neural networks. 2014.

Quoc­-Tuan Truong and Hady Lauw. Multimodal review generation for recommender systems. page 1864–1874, 2019.

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. 2017.

Jinpeng Wang, Yutai Hou, Jing Liu, Yunbo Cao, and Chin-­Yew Lin. A statistical framework for product description generation. pages 187–192, November 2017.