Learning Concept-Driven Document Embeddings for Medical Information Search

Abstract

Many medical tasks such as self-diagnosis, health-care assessment, and clinical trial patient recruitment involve the usage of information access tools. A key underlying step to achieve such tasks is the document-to-document matching which mostly fails to bridge the gap identified between raw level representations of information in documents and high-level human interpretation. In this paper, we study how to optimize the document representation by leveraging neural-based approaches to capture latent representations built upon both validated medical concepts specified in an external resource as well as the used words. We experimentally show the effectiveness of our proposed model used as a support of two different medical search tasks, namely health search and clinical search for cohorts.

Learning Unsupervised Knowledge-Enhanced Representations to Reduce the Semantic Gap in Information Retrieval

“…The results showed that constrained word representations are more effective than corpus-driven word representations when used together with bag-of-words models for re-ranking. Nguyen et al [47] present two models: the conceptual doc2vec (cdoc2vec) and the retrofitted doc2vec (rdoc2vec). Similar to the model proposed by De Vine et al [15], cdoc2vec learns document representations built upon concepts that have been previously extracted from text.…”

“…The authors employ the model in two retrieval strategies: document re-ranking and query expansion. Tamine et al [66] extend [47,48] to investigate the combined use of corpus-based information and external knowledge resources in different NLP and IR tasks. The authors compare the impact of the different learning approaches on the quality of the learned representations.…”

“…(7) conceptual doc2vec (cdoc2vec) [47] with DBOW architecture. cdoc2vec is trained over the knowledge-enhanced corpus Φ relying only on the concept vocabulary C. The query representation is the sum of its concept representations.…”

“…cdoc2vec is trained over the knowledge-enhanced corpus Φ relying only on the concept vocabulary C. The query representation is the sum of its concept representations. (8) retrofitted doc2vec (rdoc2vec) [47,66] retrofits document embeddings from doc2vec and cdoc2vec models. We derive the loss function in [47,66] to obtain the optimal (closedform) solution.…”

“…To this end, recent works integrate external knowledge into the learning process of neural language models to reduce the effect of the semantic gap between queries and documents [42,47,48]. However, even though knowledge-enhanced language models have been proven effective in many Natural Language Processing (NLP) tasks, their effectiveness is limited in IR [66].…”