Citation recommendation in scientific literature: an NLP and Deep Learning approach
Author(s)
Castelnau, Alexandre
Advisor(s)
Calhoun, Vince
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Faced with a scientific production at an ever increasing pace, the work of a researcher is accelerating and becoming more and more arduous. Thus, the writing phase of an article, an important but time-consuming phase, can appear as one of the main bottlenecks of scientific production. Wouldn't it be nice to be accompanied and helped by a tool during this time-consuming task? To have a recommendation system, able to find references that will allow us to make our work more credible and relevant?
In the following, we will study how we can face these problems and provide a model that is able to recommend references. To face this complex task, we will apply some of the latest archictectures used in NLP to the specific task of citation recommendation.
Thus, we will focus on neural networks and transformers which are currently the state of the art in NLP models. The use of our own dataset, built upon PubMed Central data, will allow us to evaluate the performance of differents methods. We also introduce new graph-based metrics for the citation recommendation task. These will also allow us to consider the results of the models under the angle of their serendipity and will allow to justify performances that go beyond the simple reproduction of already observable citation patterns.
Sponsor
Date
2023-05-02
Extent
Resource Type
Text
Resource Subtype
Thesis