Updated time: 2021-02-01 14:20:49
05T-1071206
2019-03-07
Know-how
Prior work uses a sentence’s content and contextual information to encode the dialogue or extract features, which assume a similarity between the current and nearby text. However, this assumption does not hold in conversational texts as each utterance may not be written by the same person. Considering the influence of states of each speaker on dialogue data, we propose a method which can dynamically encodes each utterance of every speaker in the dialogue. With it we incorporate the speaker’s status and their transition in conversations. When applying the proposed technique in the dialogue emotion recognition, it demonstrates a 3% performance improvement over the current state-of-the-art. In addition, it keeps its superiority on another three emotional dialogue datasets.
- We propose the dialogue embedding method that incorporates turn-taking and user state transfer and show the effectiveness of the generated embedding.
- Using these embedding, we propose the current state-of-the-art emotion recognition model for textual dialogue.
- Dialogue generation
- Dialogue emotion recognition
- Dialogue mining (for further user preference prediction)

Dialogue Embedding Flowchart. The utterance embedding is the output of this technique.
Lun-Wei Ku
If you are interested in this technology, please fill in the information on the webpage under the manufacturer option (Request for Technology Licensing, Material Transfer, or Sponsored Research) on the website of this office

Interactive Dialogue Utterance Encoding Model