Semi-supervised Abstraction-Augmented String Kernel for Multi-level Bio-Relation Extraction

Abstract

Abstract. Bio-relation extraction (bRE), an important goal in bio-text mining, involves subtasks identifying relationships between bio-entities in text at multiple levels, e.g., at the article, sentence or relation level. A key limitation of current bRE systems is that they are restricted by the availability of annotated corpora. In this work we introduce a semisupervised approach that can tackle multi-level bRE via string comparisons with mismatches in the string kernel framework. Our string kernel implements an abstraction step, which groups similar words to generate more abstract entities, which can be learnt with unlabeled data. Specifically, two unsupervised models are proposed to capture contextual (local or global) semantic similarities between words from a large unannotated corpus. This Abstraction-augmented String Kernel (ASK) allows for better generalization of patterns learned from annotated data and provides a unified framework for solving bRE with multiple degrees of detail. ASK shows effective improvements over classic string kernels on four datasets and achieves state-of-the-art bRE performance without the need for complex linguistic features.

Semi-Supervised Convolution Graph Kernels for Relation Extraction

“…That is, by using extra unlabeled data, we can consistently improve the RE performance. Table 7 shows the comparison the best performance between our methods and other methods described in [1] and [22]. We believe that our data sets are very close to those in [1] so that somehow we can directly compare with their reported results, though the preprocessing (i.e., entity chunking, splits, etc) is not identical.…”

“…The authors ran a random walk on the graph which then calculate the kernel as the sum of element-wide product between adjacency matrices of the all-path graphs. Differently from above works, recently Kuksa et al [21] treated the relation extraction task as a string classification problem using a semi-supervised string kernel approach. Word semantics patterns was also added by a semi-supervised word embedding there.…”

“…We believe that our data sets are very close to those in [1] so that somehow we can directly compare with their reported results, though the preprocessing (i.e., entity chunking, splits, etc) is not identical. For the method in [22], we tested it on our splits to guarantee a fair comparison. The method in [1] stands for the state-of-the-art performance for the graph-kernel based methods.…”

Semantic Representations for Domain Adaptation: A Case Study on the Tree Kernel-based Method for Relation Extraction

“…Sterckx et al (2014) utilize word embeddings to reduce noise of training data in distant supervision. Kuksa et al (2010) present a string kernel for bio-relation extraction with word embeddings, and Yu et al (2014; study the factor-based compositional embedding models. However, none of this work examines word embeddings for tree kernels as well as domain adaptation as we do.…”

Employing Word Representations and Regularization for Domain Adaptation of Relation Extraction

“…Kuksa et al (2010) propose an abstraction-augmented string kernel for bio-relation extraction via word embeddings. In the surge of deep learning, Socher et al (2012) and Khashabi (2013) use pre-trained word embeddings as input for Matrix-Vector Recursive Neural Networks (MV-RNN) to learn compositional structures for RE.…”