Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Facilitating Interdisciplinary Knowledge Transfer with Research Paper Recommender Systems

Eoghan Cunningham, Derek Greene, Barry Smyth

TL;DR

This work proposes a novel framework for evaluating the novelty and diversity of research paper recommendations, drawing on methods from network analysis and natural language processing, and describes a paper embedding method that provides more distant and diverse research paper recommendations without sacrificing the relevance of those recommendations compared to other state-of-the-art baselines.

Abstract

In the extensive recommender systems literature, novelty and diversity have been identified as key properties of useful recommendations. However, these properties have received limited attention in the specific sub-field of research paper recommender systems. In this work, we argue for the importance of offering novel and diverse research paper recommendations to scientists. This approach aims to reduce siloed reading, break down filter bubbles, and promote interdisciplinary research. We propose a novel framework for evaluating the novelty and diversity of research paper recommendations that leverages methods from network analysis and natural language processing. Using this framework, we show that the choice of representational method within a larger research paper recommendation system can have a measurable impact on the nature of downstream recommendations, specifically on their novelty and diversity. We highlight a novel paper embedding method, which we demonstrate offers more innovative and diverse recommendations without sacrificing precision, compared to other state-of-the-art baselines.

Facilitating Interdisciplinary Knowledge Transfer with Research Paper Recommender Systems

TL;DR

This work proposes a novel framework for evaluating the novelty and diversity of research paper recommendations, drawing on methods from network analysis and natural language processing, and describes a paper embedding method that provides more distant and diverse research paper recommendations without sacrificing the relevance of those recommendations compared to other state-of-the-art baselines.

Abstract

In the extensive recommender systems literature, novelty and diversity have been identified as key properties of useful recommendations. However, these properties have received limited attention in the specific sub-field of research paper recommender systems. In this work, we argue for the importance of offering novel and diverse research paper recommendations to scientists. This approach aims to reduce siloed reading, break down filter bubbles, and promote interdisciplinary research. We propose a novel framework for evaluating the novelty and diversity of research paper recommendations that leverages methods from network analysis and natural language processing. Using this framework, we show that the choice of representational method within a larger research paper recommendation system can have a measurable impact on the nature of downstream recommendations, specifically on their novelty and diversity. We highlight a novel paper embedding method, which we demonstrate offers more innovative and diverse recommendations without sacrificing precision, compared to other state-of-the-art baselines.
Paper Structure (16 sections, 5 equations, 3 figures, 6 tables)

This paper contains 16 sections, 5 equations, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Research Paper Recommendation
  4. Novelty and Diversity
  5. Scientific Document Representation
  6. Method: Evaluating Research Paper Recommendations
  7. Evaluating Recommendation Relevance
  8. Evaluating Recommendation Diversity and Novelty
  9. Experiment Design
  10. Data
  11. Representations for Recommendation
  12. Generating Recommendations
  13. Evaluating Recommendations
  14. Results and Discussion
  15. Conclusions
  16. ...and 1 more sections

Figures (3)

  • Figure 1: The spectrum of scientific document representation methods. Methods for paper embedding can be positioned along this spectrum, with those methods exclusively reliant on article text placed on the left, and those methods exclusively reliant on the citation graph placed on the right. Architectures that make use of both sources of information are situated between these two extremes.
  • Figure 2: A toy example designed to offer a visual demonstration of the network perspective of novelty and diversity in recommendation. The central node represents a query and the shaded nodes show examples of recommendations with different diversity and novelty scores, as measured using equations \ref{['eq:novelty']} and \ref{['eq:diversity']}, given DeepWalk embeddings as descriptions of nodes.
  • Figure 3: One layer of the ComBSAGE MPGNN framework. Unlike traditional MPGNNs, messages are not aggregated uniformly across all neighbours. The neighbours of node $v$ are grouped by the neighbourhood component function $C$. Messages from each component in $C(v)$ are aggregated separately by $\psi_i$, $\psi_2$, before combination ($\psi_3$) and update ($\phi$). For full details on the architecture, see the workshop proceedings cunningham2023graph.