Smarter Response with Proactive Suggestion: A New Generative Neural Conversation Paradigm

Authors: Rui Yan, Dongyan Zhao

IJCAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental From the experimental results, we verify the effectiveness of the new neural generative conversation paradigm. We conduct extensive experiments in a variety of humancomputer conversation setups and evaluate the performance with automatic evaluation metrics and human judgments. The experimental results are positive.
Researcher Affiliation Academia Rui Yan1,2, Dongyan Zhao1,2 1 Institute of Computer Science and Technology (ICST), Peking University 2 Beijing Institute of Big Data Research {ruiyan, zhaody}@pku.edu.cn
Pseudocode No The paper describes the Deep Dual Fusion Model using mathematical equations and prose but does not include a distinct pseudocode block or an algorithm label.
Open Source Code No The paper does not contain any statement about releasing open-source code or a link to a code repository.
Open Datasets Yes We use the data which contain a large number of human conversations from [Yao et al., 2017; Tao et al., 2018]. The data are crawled from open Web, where the users publish messages visible to the public, and then receive a bunch of subsequent replies to their utterances. We conducted the same data filtering and cleaning used in [Yan et al., 2016a].
Dataset Splits Yes We applied the validation set. All of the parameters were chosen and tuned empirically.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments, such as CPU or GPU models, or cloud computing specifications.
Software Dependencies No The paper mentions 'standard Chinese word segmentation' and 'stochastic gradient descent' but does not specify any software libraries or tools with version numbers required for reproduction.
Experiment Setup Yes We use 512-dimensional word embeddings, and they were initialized randomly. The cell units have 300 hidden units for each dimension. We used stochastic gradient descent (with a mini-batch size of 100) for optimization, gradient computed by standard back propagation. Initial learning rate was set to 0.8, and a multiplicative learning rate decay was applied.