基於性格特質的社群討論回應生成

在對話生成的研究中，雖然有部份研究針對個人化的文字生成有所探討，但主要專注於個人化的語言風格、或是職業性別等個人化的背景資訊。本研究嘗試了另一個向度的個人化文字生成，產生具有特定人格特質的文字，模擬不同性格的人，在社群媒體上的發文。本研究利用現有的資料集，再爬取社群媒體平台上的討論串，建立訓練資料集。為了強化文字生成模型對不同人格特質的建模，本研究發展了創新的鑑別學習法，引入新的損失函數，讓模型不僅能生成通順、合理的文字，並且呈現較為明顯的個人特質。實驗結果經自動與人工驗證，顯示本研究所提出之方法的效度。

關鍵字

對話生成；鑑別學習；人格特質建模

並列摘要

Previous works that attempt to emulate the human properties in dialog generation mostly focus on the incorporation of personal information or language style in the generated text. In this work, we aim to introduce a different kind of human properties in dialog generation, the personalities, to generate the response in social discussion according to a certain type of personality. We create a corpus that was crawled from a social platform with the label of personalities for the users. A novel discriminative learning approach is proposed to enhance the neural generation model toward the extrovert or the introvert personality. Both automatic and human evaluation are conducted for showing the effectiveness of our approach.

並列關鍵字

Dialog generation ； Personalities ； Discriminative learning

參考文獻

AbuShawar, B., & Atwell, E. (2015). ALICE chatbot: Trials and outputs. Computación y Sistemas, 19(4), 625-632

Google Scholar

Adiwardana, D., & Luong, T. (2020). Towards a Conversational Agent that Can Chat About… Anything. Google AI Blog.

Google Scholar

Bogatu, A., Rotarescu, D., Rebedea, T., & Ruseti, S. (2015). Conversational Agent that Models a Historical Personality. In RoCHI (pp. 81-86).

Google Scholar

Cho, K., Van Merriënboer, B., Gulcehre, C., Bahdanau, D., Bougares, F., Schwenk, H., & Bengio, Y. (2014). Learning phrase representations using RNN encoder-decoder for statistical machine translation. arXiv preprint arXiv:1406.1078..

Google Scholar

Chen, T., & Guestrin, C. (2016, August). Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining (pp. 785-794).

Google Scholar

國際替代計量

基於性格特質的社群討論回應生成

全文下載

主題瀏覽