Name
Abstract | ||
|---|---|---|
Response generation is an important component of a dialogue system. One major drawback of the existing models is that they tend to generate generic responses, and specific and latent lexical information is ignored. However, in the real world, latent lexical meaning may better reflect the user's background, which may be used to make our responses more coherent and interesting. To solve this problem, we propose a response generation model with latent lexical meaning. The model generates responses based on latent lexical meaning, which is identified by a multi-head attention mechanism, and is aligned with the output responses by lexical representation to guarantee the diversity and coherence of responses. Extensive experiments are conducted on the Sina Weibo Corpus. Experimental results show that the generated responses are evaluated as more interactive and diverse in both automatic and human evaluations. |
Year | DOI | Venue |
|---|---|---|
2020 | 10.1007/978-3-030-81197-6_41 | CHINESE LEXICAL SEMANTICS (CLSW 2020) |
Keywords | DocType | Volume |
Latent lexical meaning, Response generation, Multi-head attention | Conference | 12278 |
ISSN | Citations | PageRank |
0302-9743 | 0 | 0.34 |
References | Authors | |
0 | 5 | |