Enriching consumer health vocabulary through mining a social Q&A site: A similarity-based approach

Abstract

The widely known vocabulary gap between health consumers and healthcare professionals hinders information seeking and health dialogue of consumers on end-user health applications. The Open Access and Collaborative Consumer Health Vocabulary (OAC CHV), which contains health-related terms used by lay consumers, has been created to bridge such a gap. Specifically, the OAC CHV facilitates consumers’ health information retrieval by enabling consumer-facing health applications to translate between professional language and consumer friendly language. To keep up with the constantly evolving medical knowledge and language use, new terms need to be identified and added to the OAC CHV. User-generated content on social media, including social question and answer (social Q&A) sites, afford us an enormous opportunity in mining consumer health terms. Existing methods of identifying new consumer terms from text typically use ad-hoc lexical syntactic patterns and human review. Our study extends an existing method by extracting n-grams from a social Q&A textual corpus and representing them with a rich set of contextual and syntactic features. Using K-means clustering, our method, simiTerm, was able to identify terms that are both contextually and syntactically similar to the existing OAC CHV terms. We tested our method on social Q&A corpora on two disease domains: diabetes and cancer. Our method outperformed three baseline ranking methods. A post-hoc qualitative evaluation by human experts further validated that our method can effectively identify meaningful new consumer terms on social Q&A.

Evaluating semantic relations in neural word embeddings with biomedical and general domain knowledge bases

“…Our semantic relation evaluation dataset focused on nouns and adjectives, which was based on the statistics of synsets in WordNet [ 40 ], where 69.79% (82,115/117,659) of the synsets were nouns, and 15.43% (18,156/117,659) of synsets were adjectives. We constructed the evaluation dataset in four steps: 1) We employed a named entity recognition tool developed in our lab, simiTerm [ 41 ], to generate all the candidate evaluation terms based on the N-gram model. 2) We filtered out the noisy terms including: Terms with more than four words; Terms with a frequency < 100 in the corpus; Terms starting or ending with a stop word (We used the default stop word list in simiTerm ); Unigrams that are not noun, adjective, or gerund; Multi-grams not ending with a noun or a gerund.…”

Time-aware domain-based social influence prediction

“…Authors suggest that query expressions must be carefully chosen when sampling social media for disease-related micro-blogs. In their work in a similar domain, He et al experimented with social question-and-answer sites corpora on two disease domains -diabetes and cancer-in order to identify new, meaningful consumer terms [42]. Others developed a model comprising an ensemble of classifiers for mining social media data streams by combining similarity-based and genetic algorithm classifiers [43].…”

An Automatic Approach to Extending the Consumer Health Vocabulary

“…One example is the study by Jiang and Yang (2015) that enhanced CHV with new terms by mining online health communities and determined whether terms should be added to the ontology using co-occurrence metrics. Another example is a study by He et al (2017) exploring mining a social Q&A site to find candidate terms for expanding CHV. This stands as important work, although the results of this have not been incorporated into the publically accessible CHV.…”