Anzeige
Erschienen in:
2018 | OriginalPaper | Buchkapitel
Medical Forum Question Classification Using Deep Learning
verfasst von : Raksha Jalan, Manish Gupta, Vasudeva Varma
Erschienen in: Advances in Information Retrieval
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
Abstract
With the rapid increase in the number as well as quality of online medical forums, patients are increasingly using the Internet for health information and support. Online health forums play an important role in addressing consumers health information needs. However, given the large number of queries, and limited number of experts, a significant fraction of the questions remains unanswered. Automatic question classifiers can overcome this issue by directing questions to specific experts according to their topic preferences to get quick and better responses.
In this paper, we aim to classify health forum questions where classes of questions mainly focus on capturing user intentions. We strongly believe that a good estimate of user intentions will help direct their questions to the best responders. We propose a novel approach of combining medical domain based features with deep learning models for question classification task. To further improve performance of the data-hungry deep learning models, we resort to weak supervision strategies. We propose a new variant of the existing self-training method called “Self-Training with Lookups” for weak supervision. Our results demonstrate that combining features generated from biomedical entities along with other language representation features for deep learning networks can lead to substantial improvement in modeling user generated health content. Weak supervision further enhances the accuracy. The proposed model outperforms the state-of-the-art method on a benchmark dataset of 11000 questions with a margin of 3.13%.