A Recurrent Model for Collective Entity Linking with Adaptive Features

Authors: Xiaoling Zhou, Yukai Miao, Wei Wang, Jianbin Qin329-336

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate the performance of our proposed model on the most popular benchmark datasets for NED, and compare it with several previous state-of-the-art NED systems. 1
Researcher Affiliation Academia 1School of Computer Science and Engineering, UNSW, Australia 2College of Computer Science and Technology, DGUT, China 3Shenzhen Institute of Computer Science, Shenzhen University, China
Pseudocode No The paper describes the system architecture and its steps in prose and diagrams, but it does not contain formal pseudocode or algorithm blocks.
Open Source Code Yes Our code is released at https://github.com/tjumyk/RMA
Open Datasets Yes We validate our models on six popular benchmark datasets for NED, and compare it with several previous state-of-the-art NED systems. 1 The statistics are shown in Table 1. AIDA-Co NLL (Hoffart et al. 2011)...
Dataset Splits Yes Following previous work, we train our model on AIDA-train set, tune hyperparameters on AIDA-A set, and test on AIDA-B set (in-domain test) and other datasets (out-domain test).
Hardware Specification Yes Our model takes 10 minutes for training a local model or 15 minutes for training a global model on AIDA-train with a 10-core 3.3GHz CPU, which compares favourably with SOTA deep neural based methods, e.g. (Le and Titov 2018) takes 1.5 hours to train on the same dataset with a single Titan X GPU and (Ganea and Hofmann 2017) needs 16 hours in the same setting.
Software Dependencies No The paper mentions using 'XGBoosting' but does not specify version numbers for this or any other software dependencies.
Experiment Setup Yes In candidate entity generation, we select top-50 candidate entities from the dictionary according to the entity prior. We adopt XGBoosting with rank:pairwise objective, and set the n estimators to 4900, and max depth to 6 according to the parameter tuning on AIDA-A set, and the iteration number T for global model is 4.