Generation of topic evolution trees from heterogeneous bibliographic networks

Highlights

• Heterogeneous paper networks are useful for construction of topic evolution trees.

• Meta-path restrictions based on contribution enable context-dependent results.

• Context-sensitive topic evolution trees enhanced information retrieval results.

Abstract

The volume of the existing research literature is such it can make it difficult to find highly relevant information and to develop an understanding of how a scientific topic has evolved. Prior research on topic evolution has often leveraged refinements to Latent Dirichlet Allocation (LDA) to identify emerging topics. However, such methods do not answer the question of which studies contributed to the evolution of a topic. In this paper we show that meta-paths over a heterogeneous bibliographic network (consisting of papers, authors and venues) can be used to identify the network elements that made the greatest contributions to a topic. In particular, by adding derived edges that capture the contribution of papers, authors, and venues to a topic (using PageRank algorithm), a restricted meta-path over the bibliographic network can be used to restrict the evolution of topics to the context of interest to a researcher. We use such restricted meta-paths to construct a topic evolution tree that can provide researchers with a web-based visualization of the evolution of a scientific topic in the context of interest to them. Compared to baseline networks without restrictions, we find that restricted networks provide more useful topic evolution trees.

Introduction

An exponential growth in the output of scientific literature (Bornmann and Mutz, 2015, Price, 1961, Price, 1963, Van Raan, 2000) is one of the staples of contemporary science. In such an environment, scientists are becoming more specialized with narrower expertise (Jones, 2005, Jones, 2010), while the solutions of difficult problems in science and industry often require interdisciplinary approaches (Wagner et al., 2011) and team-based research (Falk-Krzesinski et al., 2010). Thus, ideally, scientists need to have deep knowledge in their own specialty and broad knowledge across a range of domains, i.e., be “T-shaped” (Barile, Franco, Nota, & Saviano, 2012; Donofrio, Spohrer, & Zadeh, 2010). While the Internet and electronic publications have made it easier to access an unprecedented volume and range of resources, this wealth of information can make it difficult to identify the best resources for learning the foundations of a specific topic or to identify the researchers, papers, and venues that have made the greatest contribution to a specific topic. As an example, each year the U.S. National Library of Medicine database of biomedical literature (MEDLINE) is growing by approximately 700,000 articles from 21,000 different journals and as of 2015 contains reference data on over 25 million resources.¹ While this abundance of resources provides an incredible opportunity, it can also overwhelm researchers, whether they are searching for information as part of learning a specialization or trying to develop a broad understanding of different fields.

While the current data deluge has exacerbated the problem of retrieving relevant documents, the problem itself is not new. The field of information retrieval has been proposing mostly topical solutions to the problem of retrieving relevant documents since the 1950s. At the same time, the field of bibliometrics/informetrics/scientometrics started harvesting “vast quantities of knowledge about knowledge, or metaknowledge” (Evans & Foster, 2011; p. 721) from journal articles. While the two subfields of information science mostly developed in parallel (Glänzel, 2015, Wolfram, 2015), more recently there have been efforts to bring these two subfields closer in ways that would benefit both (e.g., Glänzel, 2015, Mayr and Scharnhorst, 2015, Mutschke and Mayr, 2015, White, 2007a, White, 2007b, White, 2015, Wolfram, 2015). This paper aims to contribute to those efforts, not only by utilizing knowledge from both subfields, but by proposing solutions for identifying related research that would be useful in building information systems and delineating reference sets for informetrics research.

In this research, we propose a topic evolution tree (TET) that builds on prior research to present different evolutionary paths for topics to different individuals, based on the context that is relevant to their research. To achieve this we use a heterogeneous bibliographic network (Sun, Han, Yan, Yu, & Wu, 2011) constructed from four types of entities present in a scientific papers repository: papers (P), venues (V), authors (A), topics or keywords (K), and the relationships between them. In TET we utilize multiple meta-paths² between topic nodes in the heterogeneous graph to identify the topics that a given topic has evolved from. In constructing the TET for a topic, we calculate a score for each meta-path instance based on the edge weights of the relationships. This approach shares similarities with the calculation of path instance scores as proposed in PathSim (Sun et al., 2011). However, we propose that for topic evolution, better performance results can be obtained by adding meta-path restrictions based on a new “contribution” edge as proposed for citation recommendation (Liu, Yu, Guo, & Sun, 2014b). Namely, we use not only the edges previously used in bibliographic networks, such as “written by”, “cited by”, “published in”, or “used (topic)” (Lee & Adorna, 2012; Shi, Kong, Yu, Xie, & Wu, 2012; Sun et al., 2011;), but also contribution edges derived from PageRank calculations for each type of node (Liu et al., 2014b). For example, the “contributed-by-author” edge is calculated for each topic over a graph of authors citing other authors. We employ contribution edges as a means to restrict the context of a meta-path so that a random walk of the heterogeneous graph generates a TET showing the evolution of a topic in a particular context. In other words, we obtain the evolution of each branch of the TET based on the context or query topic that is specifically of interest to each scholar. The workflow used to generate a topic evolution tree is shown in Fig. 1.

As an example, in the information retrieval domain, a researcher might be interested in the evolution of the topic “Cloud computing”³ (user query) and the papers, authors, or venues that made the most significant contribution to the evolution of that topic. The topic “Cloud computing” would be the root node of the TET and each edge from that root to a child node represents the evolution of “Cloud computing” from a contributing topic. One of the topics contributing to “Cloud computing” is the Big Data topic “MapReduce”, so there would be an edge in the TET from “Cloud computing” to a child node labeled “MapReduce”. Since the restricted meta-path uses the contribution edge from the bibliographic network, the subtree in the TET for “MapReduce” focuses on the evolution of that topic in the context of “Cloud computing”. Thus, the evolution of the topic “MapReduce” could be different in the context of cloud computing than in the context of data security, since the papers, authors, or venues covering “MapReduce” will have made at least slightly different contributions in one context versus the other. It should be noted that this does not imply that the history of a topic such as cloud computing or MapReduce cannot be ascertained, but that the contribution edge added to a heterogeneous graph can be used to generate a TET containing the history of a topic from the viewpoint of the context of interest to the user. In the above example, the evolution of MapReduce would be included only in the context of its contribution to cloud computing. Conceptually this is similar to the research by Small, Boyack, and Klavans (2014) where the emergence of topics can be viewed from different levels of granularity.