Semi-supervised Learning for the BioNLP Gene Regulation Network

Abstract

BackgroundThe BioNLP Gene Regulation Task has attracted a diverse collection of submissions showcasing state-of-the-art systems. However, a principal challenge remains in obtaining a significant amount of recall. We argue that this is an important quality for Information Extraction tasks in this field. We propose a semi-supervised framework, leveraging a large corpus of unannotated data available to us. In this framework, the annotated data is used to find plausible candidates for positive data points, which are included in the machine learning process. As this is a method principally designed for gaining recall, we further explore additional methods to improve precision on top of this. These are: weighted regularisation in the SVM framework, and filtering out unlabelled examples based on a probabilistic rule-finding method. The latter method also allows us to add candidates for negatives from unlabelled data, a method not viable in the unfiltered approach.ResultsWe replicate one of the original participant systems, and modify it to incorporate our methods. This allows us to test the extent of our proposed methods by applying them to the GRN task data. We find a considerable improvement in recall compared to the baseline system. We also investigate the evaluation metrics and find several mechanisms explaining a bias towards precision. Furthermore, these findings uncover an intricate precision-recall interaction, depriving recall of its habitual immediacy seen in traditional machine learning set-ups.ConclusionOur contributions are twofold:1. An exploration of a novel semi-supervised pipeline. We have succeeded in employing additional knowledge through adding unannotated data points, while responding to the inherent noise of this method by imposing an automated, rule-based pre-selection step.2. A thorough analysis of the evaluation procedure in the Gene Regulation Shared Task. We have performed an in depth inquiry of the Slot Error Rate, responding to arguments that lead to some design choices of this task. We have furthermore uncovered complexities in the interplay of precision and recall that negate the customary behaviour commonplace to the machine learning engineer.

Active semi-supervised learning for biological data classification

“…The subset considered in our experiments is composed by 3, 655 scan-like samples from 53 classes (species). To describe these samples, we extract 423 features based on shape, texture and color, where the feature vector consists of: [1][2][3][4][5] scatter measures as a function of signature, [6][7][8][9][10][11][12][13] chain code histogram,…”

“…These annotation techniques usually require a standard classifier-or collection of classifiers [7,8]trained from a dataset that is fully or partially labeled. In the latter case, the semi-supervised learning approach can be used [3,[9][10][11][12][13][14][15][16], where the labeled samples (which are almost always scarce in the datasets) can propagate their labels to unlabeled samples (which represent the vast majority of them in the datasets). In the literature, it can be seen that the semi-supervised approach produces considerable improvements in the accuracy of the classifier [17][18][19][20][21].…”

Semi‐supervised long short‐term memory for human action recognition

“…One of the typical fields which is widely used is the related application of natural language processing. Provoost and Moens [22] attempt to apply semi‐supervised learning theory to the practical problem of word sense disambiguation. In processing, the need for manual tagging data is greatly reduced.…”

Artificial Intelligence and Cardiovascular Genetics

“…BCC-NER[61] NLP Bidirectional and contextual clues named entity tagger for gene/protein mention recognition BioNLP[62] …”