Augmentation, Retrieval, Generation: Event Sequence Prediction with a Three-Stage Sequence-to-Sequence Approach

Proceedings of the 29th International Conference on Computational Linguistics, pages 1865–1874 October 12–17, 2022. 1865 Augmentation, Retrieval, Generation: Event Sequence Prediction with a Three-Stage Sequence-to-Sequence Approach Bo Zhou1,2,Chenhao Wang1,2, Yubo Chen1,2, Kang Liu1,2,4, Jun Zhao1,2, Jiexin Xu3, Xiaojian Jiang3, Qiuxia Li3 1School of Artificial Intelligence, University of Chinese Academy of Sciences 2National Laboratory of Pattern Recognition, CASIA 3China Merchants Bank, 4Beijing Academy of Artificial Intelligence {bo.zhou,chenhao.wang,yubo.chen,kliu,jzhao}@nlpr.ia.ac.cn {jiexinx,jiangxiaojian,annielqx}@cmbchina.com Abstract Being able to infer possible events related to a specific target is critical to natural language processing. One challenging task in this line is event sequence prediction, which aims at predicting a sequence of events given a goal. Currently existing approach models this task as a statistical induction problem, to predict a sequence of events by exploring the simi- larity between the given goal and the known sequences of events. However, this statisti- cal based approach is complex and predicts a limited variety of events. At the same time this approach ignores the rich knowledge of external events that is important for predict- ing event sequences. In this paper, in order to predict more diverse events, we first refor- mulate the event sequence prediction problem as a sequence generation problem. Then to leverage external event knowledge, we pro- pose a three-stage model including augmen- tation, retrieval and generation. Experimen- tal results on the event sequence prediction dataset show that our model outperforms exist- ing methods, demonstrating the effectiveness of the proposed model. 1 Introduction Inferring events related to a specific target is one of the capabilities pursued by natural language under- standing, and it is also very helpful for other natural language processing (NLP) tasks, such as event ex- traction (Li et al., 2021), text summarization (Li and Zhang, 2021) and discourse understanding (Nie et al., 2019). A challenging task in this direction is event sequence prediction. Specifically, given a goal, the task aims to predict a sequence of events that fits the goal. Figure 1 shows an example of the event sequence prediction task. Given the goal Buy a mobile phone, it is expected to predict an event sequence, including four events Determine the brand, Determine the price range, Select the brand series and Pay the bill. Buy a mobile phone Determine the brand Determine the price range Select the brand series Pay the bill Event Sequence Goal Figure 1: An example of an event sequence prediction task. Given the goal Buy a mobile phone, the predicted event sequence contains four events. Although event sequence prediction is useful for many NLP tasks, it is currently understudied. Cur- rently existing approach (Zhang et al., 2020) to this task is based on statistical model that infers new sequence of events by exploiting the similarity of the given goal to known sequences of events. Al- though experiments have shown that this approach is effective, it still has two shortcomings. First, a limited variety of events are predicted. For example, when predicting the event sequence with the goal Buying a mobile phone, the method first …

Content truncated to 500 words