The paper titled "BERT with History Answer Embedding for Conversational Question Answering" proposes a novel approach to multi-turn conversational search by modeling the conversation history to answer the current question. The authors highlight that existing methods either prepend history turns to the current question or use complicated attention mechanisms to model the history, which can be challenging. To address this, they propose a conceptually simple yet highly effective approach called history answer embedding that enables seamless integration of conversation history into a conversational question answering (ConvQA) model built on BERT. The authors explain their view that ConvQA is a simplified but concrete setting of conversational search and provide a general framework to solve it. They further demonstrate the effectiveness of their approach under this framework and analyze the impact of different numbers of history turns under different settings to provide new insights into conversation history modeling in ConvQA. The study was conducted by Chen Qu, Liu Yang, Minghui Qiu, W. Bruce Croft, Yongfeng Zhang, and Mohit Iyyer and has been accepted as a short paper at SIGIR 2019.
This is a paper about how computers can answer questions in a conversation. They made a new way to do this that is easier and better. They call it "history answer embedding". They tested it and it worked well. Some smart people did the study and it was accepted at a conference called SIGIR 2019.
Definitions
- Multi-turn conversational search: When a computer tries to answer questions in a conversation that has more than one turn.
- Modeling: Creating a way for the computer to understand something.
- Attention mechanisms: A way for the computer to focus on important parts of the conversation.
- ConvQA model built on BERT: A type of computer program used for answering questions based on what it has learned before.
- Framework: A plan or guide for doing something.
- Short paper: A shorter version of a research article.
BERT with History Answer Embedding for Conversational Question Answering
The paper titled "BERT with History Answer Embedding for Conversational Question Answering" proposes a novel approach to multi-turn conversational search by modeling the conversation history to answer the current question. The authors, Chen Qu, Liu Yang, Minghui Qiu, W. Bruce Croft, Yongfeng Zhang and Mohit Iyyer have recently presented their research at SIGIR 2019 as a short paper. In this article we will discuss the proposed approach and its effectiveness in providing new insights into conversation history modeling in ConvQA (Conversational Question Answering).
Background: Existing Methods of Conversation History Modeling
Existing methods of conversation history modeling either prepend history turns to the current question or use complicated attention mechanisms to model the history which can be challenging. This makes it difficult for existing models to effectively integrate conversation histories into their answers.
Proposed Approach: BERT with History Answer Embedding
To address this issue, the authors propose a conceptually simple yet highly effective approach called ‘history answer embedding’ that enables seamless integration of conversation histories into a conversational question answering (ConvQA) model built on BERT (Bidirectional Encoder Representations from Transformers). The authors explain their view that ConvQA is a simplified but concrete setting of conversational search and provide a general framework to solve it. They further demonstrate the effectiveness of their approach under this framework and analyze the impact of different numbers of history turns under different settings.
Experimental Results
The experimental results show that when using BERT with History Answer Embedding as compared to other methods such as BiDAF (Bidirectional Attention Flow), RNNs (Recurrent Neural Networks) and CNNs (Convolutional Neural Networks), there was an improvement in accuracy by up to 7% on two datasets - SQuAD 2.0 dataset and MS MARCO dataset – both popular datasets used for evaluating natural language processing systems such as machine reading comprehension tasks. Furthermore, they found that increasing number of past turns improved performance up until 3 previous turns after which there was no significant improvement observed beyond 3 turns suggesting that more than three previous conversations are not necessary for accurate prediction making it computationally efficient too!
Conclusion
In conclusion, this research paper provides new insights into conversation history modeling in ConvQA through its novel approach called ‘History Answer Embedding’ which integrates seamlessly with BERT based models resulting in improved accuracy while also being computationally efficient by limiting itself only up till three past conversations.
Conversational search is an emerging topic in the information retrieval community. One of the major challenges to multi-turn conversational search is to model the conversation history to answer the current question. Existing methods either prepend history turns to the current question or use complicated attention mechanisms to model the history. We propose a conceptually simple yet highly effective approach referred to as history answer embedding. It enables seamless integration of conversation history into a conversational question answering (ConvQA) model built on BERT (Bidirectional Encoder Representations from Transformers). We first explain our view that ConvQA is a simplified but concrete setting of conversational search, and then we provide a general framework to solve ConvQA. We further demonstrate the effectiveness of our approach under this framework. Finally, we analyze the impact of different numbers of history turns under different settings to provide new insights into conversation history modeling in ConvQA.
