Embedding of Hierarchically Typed Knowledge Bases

Authors: Richong Zhang, Fanshuang Kong, Chenyue Wang, Yongyi Mao

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments We conduct experiments to evaluate the proposed typed models (Trans E-T and Trans H-T on binary data and m Trans H-T on multi-fold data) and their corresponding typeless models.
Researcher Affiliation Academia Richong Zhang,1 Fanshuang Kong,1 Chenyue Wang,1 Yongyi Mao2 1BDBC and SKLSDE, School of Computer Science and Engineering, Beihang University 2School of Electrical Engineering and Computer Science, University of Ottawa
Pseudocode No The paper describes the proposed scheme and models using mathematical equations and textual explanations but does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes More implementation details of the model can be found in the code on GITHUB2. 2https://github.com/kongfansh/Embedding of Hierarchically Typed KB
Open Datasets Yes Three datasets FB15K, FB15K*, and JF17K are used in the experiments. Both FB15K (Bordes et al. 2013) and JF17K (Wen et al. 2016) contain filtered data obtained from Freebase.
Dataset Splits No Table 1 lists "#instances(total/train/test)" for each dataset, indicating the data is split into training and testing sets, but a separate validation set split is not explicitly mentioned with quantities or percentages within the paper.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments, such as CPU or GPU models, memory, or cloud infrastructure specifications.
Software Dependencies No The paper mentions "SGD" (stochastic gradient descent) and refers to code on "GITHUB" but does not specify any software dependencies with version numbers, such as programming languages, libraries, or frameworks.
Experiment Setup Yes In all the models, SGD initializes all entity embedding vectors and all br vectors to random unit-length vectors. For all typed models, each ac vector is also initialized to random unit-length vectors, and each dc is initialized to 1. For m Trans H, each ar(ρ) is randomly initialized to a value in the interval (0, 1). In SGD, each mini-batch consists of 1000 packets. Each epoch loops over M/1000 batches, where M is the number of instances in the training set. For each model, SGD runs for 1000 epochs. The learning rate in the updates based on the base-cost gradients is set to 0.001, and the learning rate in the updates based on the type-cost gradient is set to 0.0003.