Learning Triple Embeddings from Knowledge Graphs

Authors: Valeria Fionda, Giuseppe Pirrò3874-3881

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

Reproducibility Variable Result LLM Response
Research Type Experimental We evaluate our approach on different realworld knowledge graphs and compared it with related work. We also show an application of triple embeddings in the context of user-item recommendations. We now report on the evaluation and comparison with related work. Triple2Vec has been implemented in Python and uses the Gensim1 library to learn embeddings.
Researcher Affiliation Academia 1Department of Mathematics and Computer Science, University of Calabria Via Pietro Bucci 30B, 87036, Rende (CS), Italy 2Department of Computer Science, Sapienza University of Rome Via Salaria 113, 00198, Rome, Italy fionda@mat.unical.it, pirro@di.uniroma1.it
Pseudocode Yes Algorithm 1: Build Triple Line Graph (G)
Open Source Code No The paper mentions using and comparing against existing open-source implementations (Gensim, KPRN) but does not provide a link or explicit statement for its own Triple2Vec source code.
Open Datasets Yes In the experiments, we used three real-world data sets. DBLP (Huang and Mamoulis 2017) about authors, papers, venues, and topics. This dataset has 16K nodes, 52K edges, and 4 predicate types. [...] Foursquare (Hussein, Yang, and Cudr e-Mauroux 2018) with 30K nodes and 83K edges [...] Finally, we used a subset of Yago (Huang and Mamoulis 2017) in the domain of movies.
Dataset Splits No The paper mentions varying the percentage of training data for evaluation but does not explicitly state the use or size of a separate validation set for model tuning or early stopping.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions 'Python', 'Gensim', and 'pykg2vec' but does not specify their version numbers, which is required for reproducible software dependencies.
Experiment Setup Yes For sake of space, in what follows, we only report the best results obtained by setting the parameters of Triple2Vec as follows: number of walks per node n=10, maximum walk length L = 100, window size (necessary for the context in the Skip-gram model) w = 10. Moreover, we used d=128 as a dimension of the embeddings. The number of negative samples Γ is set to 50. All results are the average of 10 runs.