Learning the Compositional Visual Coherence for Complementary Recommendations

Authors: Zhi Li, Bo Wu, Qi Liu, Likang Wu, Hongke Zhao, Tao Mei

IJCAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on the large-scale real-world data clearly demonstrate the effectiveness of CANN compared with several state-of-the-art methods.
Researcher Affiliation Collaboration Zhi Li1 , Bo Wu2 , Qi Liu1,3, , Likang Wu3 , Hongke Zhao4 and Tao Mei5 1Anhui Province Key Laboratory of Big Data Analysis and Application, School of Data Science, University of Science and Technology of China 2Columbia University 3School of Computer Science and Technology, University of Science and Technology of China 4The College of Management and Economics, Tianjin University 5JD AI Research
Pseudocode Yes Algorithm 1 Compositional Optimization Strategy
Open Source Code Yes The datasets and source codes are available in our project pages 1. 1https://data.bdaa.pro/BDAA Fashion/index.html
Open Datasets Yes We evaluate our proposed method on a real-world dataset, i.e., Polyvore dataset [Han et al., 2017; Vasileva et al., 2018].
Dataset Splits Yes Then, we split the dataset into 59,212 outfits with 221,711 fashion items for training, 3,000 outfits for validation and 10,218 outfits for testing.
Hardware Specification Yes Our model and all the compared methods are developed and trained on a Linux server with two 2.20 GHz Intel Xeon E52650 v4 CPUs and four TITAN Xp GPUs.
Software Dependencies No The paper mentions software components and models like "Google Net Inception V3 model" and "Re LU function", but it does not provide specific version numbers for any software, libraries, or frameworks used.
Experiment Setup Yes The number of visual space is set to S = 4 and for each visual space, we set ds = df/S = 128 and b = 4 unless otherwise noted. Our model is trained with an initial learning rate of 0.2 and is decayed by a factor of 2 every 2 epochs. The batch size is set to 9, seed collection length k is set to 8.