LENA: Locality-Expanded Neural Embedding for Knowledge Base Completion

Authors: Fanshuang Kong, Richong Zhang, Yongyi Mao, Ting Deng2895-2902

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental This model is tested on four standard datasets and compared with several stateof-the-art models for knowledge base completion. Extensive experiments suggest that LENA outperforms the existing models in virtually every metric.
Researcher Affiliation Academia 1SKLSDE, School of Computer Science and Engineering, Beihang University, Beijing, China 2Beijing Advanced Institution on Big Data and Brain Computing, Beihang University, Beijing, China 3School of Electrical Engineering and Computer Science, University of Ottawa, Ottawa, Canada
Pseudocode No The paper describes its methods using mathematical equations and textual descriptions, but it does not contain a clearly labeled pseudocode block or algorithm.
Open Source Code Yes Our code is at https://github.com/fskong/LENA.
Open Datasets Yes FB15K, WN18, FB15K-237 and WN18-RR are the most commonly used dataset for KB link prediction tasks. We conduct empirical studies on these four datasets to evaluate our model for link prediction. All four datasets represent multi-relational data as triples. FB15K is a subset of Free Base, a large-scale general-fact KB, and WN18 is a subset of Word Net, in which entities represent word senses and relations describe lexical relationships between two word senses.
Dataset Splits Yes The statistics of the datasets are listed as in Table 1. Datasets entities relations triples(train/test/valid) FB15K 14,951 1,345 483,142 / 59,071 / 50,000 WN18 40,943 18 141,442 / 5,000 / 5,000 FB15K-237 14,541 237 272,115 / 20,466 / 17,535 WN18-RR 40,943 11 86,835 / 3,134 / 3,034
Hardware Specification No The paper mentions 'on the same computer' but does not provide specific hardware details such as CPU model, GPU model, or memory specifications used for the experiments.
Software Dependencies No The paper mentions optimizers like Adam and various models but does not provide specific version numbers for software dependencies such as programming languages, libraries, or frameworks (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes For LENA, the values chosen for retention rate δ are {0.1, 0.25, 0.5}. Optimizer Adam (Kingma and Ba 2014) with initial learning rate 0.01 and mini-batch size 200 is run for 30 epochs. Embedding dimension is chosen as 200. Other hyper-parameter settings of LENA are listed in Table 2.