SurfCon

Abstract

Unstructured clinical texts contain rich health-related information. To better utilize the knowledge buried in clinical texts, discovering synonyms for a medical query term has become an important task. Recent automatic synonym discovery methods leveraging raw text information have been developed. However, to preserve patient privacy and security, it is usually quite difficult to get access to largescale raw clinical texts. In this paper, we study a new setting named synonym discovery on privacy-aware clinical data (i.e., medical terms extracted from the clinical texts and their aggregated co-occurrence counts, without raw clinical texts). To solve the problem, we propose a new framework SurfCon that leverages two important types of information in the privacy-aware clinical data, i.e., the surface form information, and the global context information for synonym discovery. In particular, the surface form module enables us to detect synonyms that look similar while the global context module plays a complementary role to discover synonyms that are semantically similar but in different surface forms, and both allow us to deal with the OOV query issue (i.e., when the query is not found in the given data). We conduct extensive experiments and case studies on publicly available privacy-aware clinical data, and show that SurfCon can outperform strong baseline methods by large margins under various settings.

A Visually Enhanced Neural Encoder for Synset Induction

“…In other words, only a collection of tags is given for clustering identical semantics. To this end, these methods can roughly be grouped into two categories: corpus statistics and pattern-based methods [7][8][9][10][11] and distributional representation-based methods [13][14][15][16].…”

“…However, previous methods focus on the patterns, which cannot effectively capture the semantics of tag words. In order to alleviate the problem, inspired by the word2vec methods [19,32], the distributional representations have been incorporated for the synset induction task [13][14][15]. Mamou et al [13] presented an endto-end workflow to induce synsets, which were based on multi-context word embedding.…”

“…The learned semantics were then used to mine entity synsets. Wang et al [15] proposed a SurfCon method to compute the semantic similarity between words, in which the surface form, and global context information was used.…”

“…Given a collection of tags, the synset induction algorithms aim to cluster the tags such that each cluster refers to identical semantics. These methods can roughly be grouped into two categories: corpus statistics and patterns based [7][8][9][10][11][12] and distributional representation based [13][14][15][16]. Those methods generally achieve a promising performance.…”