This repository contains the code for our SETN 2020 paper with title: Zero-Shot Classification of Biomedical Articles with Emerging MeSH Descriptors
In brief, we develop a Zero-shot learning (ZSL) mechanism (ZSLbioSentMax) that tries to perform on-the-fly classification examining text-raw data that stem from PubMed abstracts searching for emerging MeSH terms that appear each upcoming year. Therefore, bioBERT embeddings are exploited computed on sentence-level so as to exploit better the semantic relationships that occur into the corresponding abstracts, avoiding misclassification that may occur in case of word-guided approaches.
Later, we use the same mechanism for providing weak labels into gathered data from previous years, and then apply two different Weakly supervised learning (WSL) approaches for examining their performance on the same tasks: i) WSLbioSentMax(bioBERT), ii) WSLbioSentMax(tfidf).
Three different MeSH terms were investigated here, under single-label mode. Appropriate baselines and a state-of-the-art approach have been developed as they are described into the original paper. An expansion of our work towards multi-label classification tasks covering more complex cases and larger number of labels is the next step in the context of a larger project, called AMULET (more info on Additional resources section).
ZSL_and_WSL_implementations_setn2020.py file contains a proper menu for selecting any of the implemented approaches, regarding the datasets inside raw data folder. In case of additional desired MeSH terms, you can add .txt files where each instance is in the next format:
b'X labels: #y1#y2#...#yn' (X: abstract raw-text, y1,y2,...,yn the corresponding labels).
The results are placed in folder with compatible names, while additional plots or code scripts for executing optimization of the existing hyper-parameters are placed accordingly.
Hint: Mode 6 and Mode 7 have been added so as to save the appropriate embeddings per MeSH term, for applying the optimization based on separate threshold values. Otherwise, each run would produce results only for one specific threshold value, increasing highly the spent computational resources.
If you find the code useful for your research, please cite our paper:
@inproceedings{DBLP:conf/setn/MylonasKT20,
author = {Nikolaos Mylonas and
Stamatis Karlos and
Grigorios Tsoumakas},
editor = {Constantine D. Spyropoulos and
Iraklis Varlamis and
Ion Androutsopoulos and
Prodromos Malakasiotis},
title = {Zero-Shot Classification of Biomedical Articles with Emerging MeSH
Descriptors},
booktitle = {{SETN} 2020: 11th Hellenic Conference on Artificial Intelligence,
Athens, Greece, September 2-4, 2020},
pages = {175--184},
publisher = {{ACM}},
year = {2020},
url = {https://dl.acm.org/doi/10.1145/3411408.3411414},
timestamp = {Mon, 21 Sep 2020 14:53:26 +0200},
biburl = {https://dblp.org/rec/conf/setn/MylonasKT20.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
Our code has been tested on Windows10 using python 3.7.6. The mentioned time responses correspond to a working station embedded with Intel Core i7-9700 (3GHz) processor and 32 GB RAM. The next libaries are necessary:
- Numpy
- bioBERT
- Spacy
- Pandas
- Seaborn and Matplotlib (for graphing)
| Name (English/Greek) | |
|---|---|
| Nikolaos Mylonas (Νικόλαος Μυλωνάς) | [email protected] |
| Stamatis Karlos (Σταμάτης Κάρλος) | [email protected] |
| Grigorios Tsoumakas (Γρηγόριος Τσουμάκας) | [email protected] |
The research work was supported by the Hellenic Foundation forResearch and Innovation (H.F.R.I.) under the “First Call for H.F.R.I.Research Projects to support Faculty members and Researchers and the procurement of high-cost research equipment grant” (ProjectNumber: 514).
