The Style-Content Duality of Attractiveness: Learning to Write Eye-Catching Headlines via Disentanglement

Authors: Mingzhe Li, Xiuying Chen, Min Yang, Shen Gao, Dongyan Zhao, Rui Yan13252-13260

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on the public Kuaibao dataset show that DAHG achieves state-of-the-art performance.
Researcher Affiliation Academia 1 Wangxuan Institute of Computer Technology, Peking University,Beijing,China 2 Center for Data Science, AAIS, Peking University,Beijing,China 3 Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China 4 Gaoling School of Artificial Intelligence, Renmin University of China 5 Beijing Academy of Artificial Intelligence
Pseudocode No No structured pseudocode or algorithm blocks (e.g., labeled 'Algorithm' or 'Pseudocode') were found.
Open Source Code No The paper does not provide an explicit statement about releasing source code or a link to a code repository for the described methodology.
Open Datasets Yes We use the public Tencent Kuaibao Chinese News dataset1 proposed by Qin et al. (2018). The dataset contains 160,922 training samples, 1,000 validation samples, and 1,378 test samples. 1https://kuaibao.qq.com/
Dataset Splits Yes The dataset contains 160,922 training samples, 1,000 validation samples, and 1,378 test samples.
Hardware Specification Yes Our experiments are implemented in Tensorflow (Abadi et al. 2016) on an NVIDIA GTX 1080 Ti GPU.
Software Dependencies No The paper mentions 'Tensorflow' but does not provide a specific version number for it or for any other software libraries or dependencies.
Experiment Setup Yes Experiments are performed with a batch size of 64. We pad or cut the input document to 400 words and the prototype headline to 30 words. The maximum decode step is set to 30, and the minimum to 10 words. We initialize all of the parameters in the model using a Gaussian distribution. We choose Adam optimizer for training, and use dropout in the VAE encoder with keep probability as 0.8. For testing, we use beam search with size 4 to generate a better headline.