Neural personalized response generation as domain adaptation

By Weinan Zhang, Ting-Wei Liu, Yifa Wang, Qingfu Zhu. 2017

109 citations recorded by the source corpus.

One of the most crucial problem on training personalized response generation models for conversational robots is the lack of large scale personal conversation data. To address the problem, we propose a two-phase approach, namely initialization then adaptation, to first pre-train an optimized RNN encoder-decoder model (LTS model) in a large scale conversational data for general response generation and then fine-tune the model in a small scale personal conversation data to generate personalized responses. For evaluation, we propose a novel human aided method, which can be seen as a quasi-Turing test, to evaluate the performance of the personalized response generation models. Experimental results show that the proposed personalized response generation model outperforms the state-of-the-art approaches to language model personalization and persona-based neural conversation generation on the automatic evaluation, offline human judgment and the quasi-Turing test.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Challenges in Building Intelligent Open-domain Dialog Systems (2019, methodology): “[176] used an embedding vector to represent a user (speaker) persona and fed the user embedding into each decoding position of the decoder.”
A Survey on Dialogue Systems: Recent Advances and New Frontiers (2017, methodology): “Since the training data comes from different speakers with inconsistency, [106] proposed a two-phase training approach which initialized the model using large scale data and then fine-tuned the model to generate personalized response.”
Personalized Response Generation via Domain adaptation (2017, methodology): “Recently, several studies have been proposed to capture personal characteristics and handle the information consistency of a user [2, 5, 11, 12, 14].”
Condition Aware and Revise Transformer for Question Answering (2020, background): “Yang et al. [33] and Zhang et al. [36] attempt to introduce personalized information to dialogs by transfer learning.”
I like fish, especially dolphins: Addressing Contradictions in Dialogue Modeling (2020, background): “…embeddings (Li et al., 2016; Zhang et al., 2019a), neural long-term memories (Bang et al., 2015), hierarchical neural architecture (Serban et al., 2016), latent variables (Ser-ban et al., 2017), topical attention (Dziri et al., 2019b), or even self-learned feature vectors (Zhang et al., 2019b).”
How Should My Chatbot Interact? A Survey on Social Characteristics in Human–Chatbot Interaction Design (2019, background): “the expected outcome. Although the literature in computational linguistics has proposed algorithms and statistical models to manipulate language style and matching (see e.g., [Prabhumoye et al. 2018; Zhang et al. 2018b]), to the best of our knowledge, these strategies have not been evaluated in the ”
Personalizing a Dialogue System With Transfer Reinforcement Learning (2016, background): “(Zhang et al. 2017) study the personalized response generation problem in single-turn open-domain dialogue system, which does not model the multi-turn dialogue context or the dialogue state.”