Learning Conceptual-Contextual Embeddings for Medical Text
Authors: Xiao Zhang, Dejing Dou, Ji Wu9579-9586
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on electronic health records (EHRs) and medical text processing benchmarks showed our model gives a major boost to the performance of supervised medical NLP tasks. |
| Researcher Affiliation | Collaboration | Xiao Zhang,1 Dejing Dou,3,4 Ji Wu1,2 1Department of Electronic Engineering, Tsinghua University 2Institute for Precision Medicine, Tsinghua University 3Department of Computer and Information Science, University of Oregon 4Baidu Research |
| Pseudocode | Yes | Algorithm 1 Training CC embedding model |
| Open Source Code | No | The paper does not contain any statement about making the source code available or provide a link to a code repository for the methodology described. |
| Open Datasets | Yes | To encode the knowledge into a text representation model, we use knowledge graph embedding task, like in (Bordes et al. 2013). To show our approach is scalable to large knowledge graphs in the medical domain, we use the UMLS database (Bodenreider 2004) for learning to encode medical concept embeddings. Context corpus. Learning to recognize concepts in text requires seeing concept names in context. We prepared a corpus from Pub Med citations and MIMIC-III critical care database (Johnson et al. 2016). |
| Dataset Splits | Yes | The patients are split into training (80%), validation (10%) and testing (10%) sets with 5-fold cross validation. Original dataset contains 11232 sentence pairs for training and 1395 and 1422 pairs for development and testing. |
| Hardware Specification | No | The paper does not provide specific hardware details such as exact GPU/CPU models, processor types, or memory amounts used for running its experiments. |
| Software Dependencies | No | The paper mentions software components like 'Bio Word Vec' and 'Apache Solr TM' but does not provide specific version numbers for these or other software dependencies. |
| Experiment Setup | Yes | In experiments we use 200-dimensional word embeddings and 2-layer bi-directional LSTM network with also 200 dimensions. In the ranking loss L, Euclidean norm is used in distance function d and margin γ = 0.1. Vanilla stochastic gradient descent with learning rate l = 1.0 is used to optimize the network. A total of 10 epochs is trained on 23 million training triplets. |