A Sentiment-Controllable Topic-to-Essay Generator with Topic Knowledge Graph

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 3336–3344 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 3336 A Sentiment-Controllable Topic-to-Essay Generator with Topic Knowledge Graph Lin Qiao1∗, Jianhao Yan2, Fandong Meng2, Zhendong Yang, and Jie Zhou2 1School of Software and Microelectronics, Peking University 2Pattern Recognition Center, WeChat AI, Tencent Inc, Beijing, China qiaolin66666@gmail.com {elliottyan,fandongmeng,withtomzhou}@tencent.com Abstract Generating a vivid, novel, and diverse es- say with only several given topic words is a challenging task of natural language genera- tion. In previous work, there are two problems left unsolved: neglect of sentiment beneath the text and insufficient utilization of topic- related knowledge. Therefore, we propose a novel Sentiment-Controllable topic-to-essay generator with a Topic Knowledge Graph enhanced decoder, named SCTKG, which is based on the conditional variational auto- encoder (CVAE) framework. We firstly in- ject the sentiment information into the gen- erator for controlling sentiment for each sen- tence, which leads to various generated essays. Then we design a Topic Knowledge Graph en- hanced decoder. Unlike existing models that use knowledge entities separately, our model treats knowledge graph as a whole and en- codes more structured, connected semantic in- formation in the graph to generate a more rel- evant essay. Experimental results show that our SCTKG can generate sentiment control- lable essays and outperform the state-of-the- art approach in terms of topic relevance, flu- ency, and diversity on both automatic and hu- man evaluation. 1 Introduction Topic-to-essay generation (TEG) task aims at gen- erating human-like paragraph-level texts with only several given topics. It has plenty of practical appli- cations, e.g., automatic advertisement generation, intelligent education, or assisting in keyword-based news writing (Lepp¨anen et al., 2017). Because of its great potential in practical use and scientific re- search, TEG has attracted a lot of interest. (Feng et al., 2018; Yang et al., 2019). However, In TEG, two problems are left to be solved: the neglect of sentiment beneath the text and the insufficient utilization of topic-related knowledge. ∗This work is done when Lin Qiao was interning at Pattern Recognition Center, WeChat AI, Tencent Inc, China Love is a kind of emotion. Love experience is important to every one.... It’s been half a year since I fell in love with my boyfriend. He treats me very well...... It‘s been half a year since I fell in love with my boyfriend. But these few months my boyfriend rarely contacted me....... I get addicted to smoking after broke up with him. I hope someone can comfort and encourage me...... I get addicted to smoking after broke up with him. I feel sick and headache after smoking every time...... positive positive positive negative negative positive negative negative without sentiment Love Emotion Experience Love Emotion Experience Input topics Output essays Figure 1: Examples of comparison between the gener- ated essays with sentiment control and without senti- ment. We show the first two sentences for each gener- ated essay and denote positive sentences in red and neg- ative sentences in blue. Sentences without sentiment label …

Content truncated to 500 words