Voices in Cyberspace: A Move-Step Analysis of Politics-centered Online News Editorials

Question-Answer Pairing from IM Conversations via Message Merging and Reply-to Prediction Thamolwan Poopradubsil Department of Computer Science Kasetsart University, Thailand tmw.poopradubsil@gmail.com Chia-Hui Chang Department of Computer Science and Information Engineering National Central University, Taiwan chia@csie.ncu.edu.tw Abstract Preparing question-answer pairs from conver- sation logs (chat logs) is often considered a prerequisite for downstream dialogue tasks such as response generation and response se- lection tasks. In this paper, we study a task called reply-to prediction, which can be used to prepare question-answer pairs. Unlike other works, our data comes from the instant mes- saging (IM) platform where participants could split long sentences into short utterances and send them in multiple messages. We consider a task called message merging task which aims to determine whether those messages need to be merged or not before generating message pairs for reply-to prediction task. The theory behind this task is similar to, yet differ- ent from reply-to prediction task in which this task uses the messages from the same speaker to predict whether these two messages are re- lated or not. We propose a CONTEXT-AOA model to include the context (previous dia- logue) as additional input apart from pairwise messages. Our experiments show that our proposed model outperforms both single-turn (pairwise) conversation models and multi-turn (context-aware) conversation models on mes- sage merging task and achieves a close per- formance compares to other multi-turn mod- els on reply-to prediction for manually labeled data and outperforms other models when us- ing heuristic labeled data. 1 Introduction Preparation of questions-answers from conversation logs (also known as chat logs) between customers and enterprise personnel is important for the devel- opment of chatbots. For example, Figure 1 shows a conversation between a customer service staff and a client. The goal here is to find the question-answer pairs such as (d, e), (p, q), from the chat logs such that the question-answer pairs can be used as the chatbot’s responses to the corresponding question. Finding the question-answer pairs is similar to the link prediction task where the goal is to de- cide whether an incoming message replies to the existing question based on the similarity between messages. Link prediction can be defined either as a reply-to prediction or as a response selection. Both of the problems require message pair similari- ties to determine which previous or following mes- sage is the best-matched candidate question or re- sponse. Similar to conversation disentanglement, earlier works on response selection only use the last message in the context for matching with the can- didate response message from different utterances (Wang et al., 2013). However, human does not give the response entirely based on a link prediction task. Real-world chat logs are multi-turn conversations, thus including the context as another input is nec- essary in order to allow a model to remember what has been discussed so far. Recent works show that including the multi-turn conversation improves the response selection’s performance, e.g. (Zhou et al., 2016; Zhou et al., 2018; Wu et al., 2019). In this work, …

Content truncated to 500 words