Context-aware Path Ranking for Knowledge Base Completion

Authors: Sahisnu Mazumder, Bing Liu

IJCAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on three large KBs show that the path features (fewer in number) discovered by C-PR not only improve predictive performance but also are more interpretable than existing baselines.
Researcher Affiliation Academia Sahisnu Mazumder, Bing Liu Department of Computer Science, University of Illinois at Chicago, USA sahisnumazumder@gmail.com, liub@cs.uic.edu
Pseudocode No The paper describes the algorithm steps in text and provides an example trace in a table, but it does not include a formal pseudocode block or algorithm listing.
Open Source Code No The paper does not provide any statement or link indicating that the source code for their method is openly available.
Open Datasets Yes We used three datasets for our experiments: two standard ones, viz (1) FB15k 4, (2) Word Net 4 and (3) one dataset created by us from Concept Net 5 triples. FBK15k is a relatively dense subgraph of Freebase. The Word Net dataset is comparatively small and Concept Net dataset is the largest one among the three.
Dataset Splits Yes For each of the 25 relations, we randomly shuffle the list of distinct triples, choose 1000 triples and split them into 80% training, 20% test.
Hardware Specification No We performed experiments on Amazon EC2 linux server having Intel Xeon processor, 264GB RAM and 64 CPU cores in order to support the baselines that use exhaustive path search.
Software Dependencies No We use sklearn python ML library for training the LR model with L2-regularization.
Experiment Setup Yes We use sklearn python ML library for training the LR model with L2-regularization. Other parameters of the LR model are: tolerance set to 0.0001, maximum iterations for convergence to 200, and class weight set as balanced . We choose LR because LR not only has been used in existing PR-based approaches [Lao and Cohen, 2010; Gardner et al., 2014; Gardner and Mitchell, 2015], but also has been shown to give better performances compared to SVM [Wang et al., 2016].