Dataset Recommendation via Variational Graph Autoencoder

Abstract

This paper targets on designing a query-based dataset recommendation system, which accepts a query denoting a user's research interest as a set of research papers and returns a list of recommended datasets that are ranked by the potential usefulness for the user's research need. The motivation of building such a system is to save users from spending time on heavy literature review work to find usable datasets. We start by constructing a two-layer network: one layer of citation network, and the other layer of datasets, connected to the firstlayer papers in which they were used. A query highlights a set of papers in the citation layer. However, answering the query as a naive retrieval of datasets linked with these highlighted papers excludes other semantically relevant datasets, which widely exist several hops away from the queried papers. We propose to learn representations of research papers and datasets in the two-layer network using heterogeneous variational graph autoencoder, and then compute the relevance of the query to the dataset candidates based on the learned representations. Our ranked datasets shown in extensive evaluation results are validated to be more truly relevant than those obtained by naive retrieval methods and adoptions of existing related solutions.

Scientific Dataset Discovery via Topic-level Recommendation

“…Deepwalk [20], Node2vec [39], Metapath2vec [44]. (d) Network and Content Representation Learning Techniques: Text-associated Deepwalk (TADW) [45], NRL+ LSI [46], NRL+ Doc2vec [47], GCN [48], HVGAE [6]. (e) Topic Modeling based Techniques: HFT [23], CDL [30], CVAE-CF [49], JMARS [29], CTFP [28].…”

“…These approaches satisfy the needs of a specific domain and are inapplicable to our studied problem. In our study, we need to jointly model the relationship between homogeneous nodes (paper) and the relationship between heterogeneous nodes (paper and dataset), also the text information of nodes (paper) for retrieving datasets based on the query specified as a list of papers as studied in [6]. However, our work is different from [6] that learn paper and dataset representation via Variational Autoencoders, as we model the shared topics between paper and dataset nodes, to help us understand paper-paper citation and paper-dataset citation as well as to facilitate the query based dataset recommendation.…”

“…Second, the existing methods allow researchers to find scientific datasets by (a) reading a mass of relevant papers or (b) using search engines such as Google Dataset search 1 . However, the former method is a timeconsuming process, and the latter method usually requires to refine the query words repeatedly after checking the search results, which may be difficult for entry-level researchers due to their shallow understanding about their research problem [6]. Both these methods fail to consider the common semantics between user's research interests and available datasets.…”

Recommending scientific datasets using author networks in ensemble methods

“…Scientists are encouraged by funding agencies to publish datasets using the FAIR principles [41]. It is widely acknowledged that open and FAIR datasets contribute to both the transparency of science, to its quality, its reproducibility Because of the increasing importance of open datasets for modern science, a number of dataset search engines can be found online nowadays, including Google Dataset Search, 1 Mendeley Data, 2 Microsoft Rearch Open Data 3 and others. These dataset search engines help researchers to find datasets based on an input query consisting of keywords.…”

“…Ellefi et al [11] provide a dataset recommendation approach by considering the overlap between the schema of two datasets, which achieved perfect recall and a precision of 0.53. Altaf et al [1] provide a dataset recommendation method based on a set of research papers given by the user, achieving 0.92 recall score and 0.18 precision score. Giseli et al [28] present two approaches for dataset recommendation, based on Bayesian classifiers and on Social Network connections, which achieved a mean average precision score of around 0.6.…”