Advisor BERT: Building a search engine to help students in choosing a scientific advisor

Abstract

Choosing a laboratory for PhD and postdoctoral study is a common challenge among students. Often, the final decision is mainly influenced by prestige or available connections, not students’ interests. This is one of the factors that explain the high Ph.D. dropout rate. 

To help students in making this decision, we have developed a special-purpose academic search engine. The application allows a student to input any specific topic they are interested in and get a map with all of the active researchers who produce work on the topic. Researchers are ranked by their impact index and links to their social media are provided. This allows students to broaden their scope of researchers working in their field of interest and find potential collaborators or advisors. 

The tool uses data from the OpenAlex database, an open-source catalogue of research papers, researchers and their affiliations. We use similarity-based techniques with a pre-trained neural transformer language model to search through relevant publications’ abstracts and then rank their authors. We then use the data from OpenAlex to map the researcher and their affiliations and display the results with a React-Flask web application. 

We have developed a working prototype with a dataset of 800 thousand papers. It has been tested on a group of bachelor students. We have received feedback that it allowed them to find relevant researchers in the field of AI. Now we are planning to conduct a systematic study on a larger sample of students.