Combining Impression Feature Representation for Multi-turn Conversational Question Answering

Combining Impression Feature Representation for Multi-turn Conversational Question Answering Shaoling Jing1,2,3,* and Shibo Hong2 and Dongyan Zhao1 and Haihua Xie2 and Zhi Tang1 1. Wangxuan Institute of Computer Technology, Peking University, Beijing, China 2. State Key Laboratory of Digital Publishing Technology, Peking University Founder Group Co. LTD., Beijing, China 3. Postdoctoral Workstation of the Zhongguancun Haidian Science Park, Beijing, China {jingshaoling, zhaody, tangzhi}@pku.edu.cn {hongshibo, xiehh}@founder.com Abstract Multi-turn conversational Question Answering (ConvQA) is a practical task that requires the understanding of conversation history, such as previous QA pairs, the passage context, and cur- rent question. It can be applied to a variety of scenarios with human-machine dialogue. The major challenge of this task is to require the model to consider the relevant conversation history while understanding the passage. Existing methods usually simply prepend the history to the cur- rent question, or use the complicated mechanism to model the history. This article proposes an impression feature, which use the word-level inter attention mechanism to learn multi-oriented information from conversation history to the input sequence, including attention from history tokens to each token of the input sequence, and history turn inter attention from different history turns to each token of the input sequence, and self-attention within input sequence, where the input sequence contains a current question and a passage. Then a feature selection method is designed to enhance the useful history turns of conversation and weaken the unnecessary infor- mation. Finally, we demonstrate the effectiveness of the proposed method on the QuAC dataset, analyze the impact of different feature selection methods, and verify the validity and reliability of the proposed features through visualization and human evaluation. 1 Introduction Conversational Question Answering (ConvQA) is a new question answering task that requires a compre- hension of the context, which has recently received more and more attention (Zhu et al., 2018; Qu et al., 2019a; Qu et al., 2019b; Meng et al., 2019; Pruthi et al., 2020). Since conversation is one of the most natural ways for humans to seek information, it carries over context through the dialogue flow. Specifi- cally, we ask other people a question, dependending on their answer, we follow up with a new question, and second answer with additional information will be given based on what has been discussed (Reddy et al., 2019). Therefore, multi-turn conversational question answering is formed in this way. It can be used in many fields as a personal assistant systems, such as, customer service, medical, finance, ed- ucation, etc. Moreover, with the rapid development of artificial intelligence technology in theory and practical applications, many personal assistant products have been launched in the market, such as Al- ibaba AliMe, Apple Siri, Amazon Alexa, etc. Although these assistants are capable to cover some simple tasks, they cannot handle complicated information-seeking conversations that require multiple turns of interaction (Qu et al., 2019b). In the tasks of two recent multi-turn ConvQA datasets, CoQA (Reddy et al., 2019) and QuAC (Choi et al., 2018), given a passage, a question, and the conversation …

Content truncated to 500 words