Leveraging Capsule Routing to Associate Knowledge with Medical Literature Hierarchically

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 3518–3532 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 3518 Leveraging Capsule Routing to Associate Knowledge with Medical Literature Hierarchically Xin Liu1,2, Qingcai Chen1,2,∗, Junying Chen1,2, Wenxiu Zhou1,2, Tingyu Liu1, Xinlan Yang1 and Weihua Peng3 1Harbin Institute of Technology, Shenzhen, 2Peng Cheng Laboratory, 3Baidu International Technology (Shenzhen) Co., Ltd {hit.liuxin,junying.chen.cs}@gmail.com, qingcai.chen@hit.edu.cn wen.xiu.zhou@outlook.com, {liu-ty,xinlan.y}@foxmail.com, pengweihua@baidu.com Abstract Integrating knowledge into text is a promis- ing way to enrich text representation, espe- cially in the medical field. However, un- differentiated knowledge not only confuses the text representation but also imports un- expected noises. In this paper, to alleviate this problem, we propose leveraging capsule routing to associate knowledge with medical literature hierarchically (called HiCapsRKL). Firstly, HiCapsRKL extracts two empirically designed text fragments from medical litera- ture and encodes them into fragment repre- sentations respectively. Secondly, the capsule routing algorithm is applied to two fragment representations. Through the capsule com- puting and dynamic routing, each represen- tation is processed into a new representation (denoted as caps-representation), and we inte- grate the caps-representations as information gain to associate knowledge with medical liter- ature hierarchically. Finally, HiCapsRKL are validated on relevance prediction and medical literature retrieval test sets. The experimen- tal results and analyses show that HiCapsRKL can more accurately associate knowledge with medical literature than the mainstream meth- ods. In summary, HiCapsRKL can efficiently help selecting the most relevant knowledge to the medical literature, which may be an al- ternative attempt to improve knowledge-based text representation. Source code is released on GitHub 1. 1 Introduction Knowledge is known as a triple to describe the rela- tionship (r) between head entity (eh) and tail entity (et) with the format of <eh, r, et>. The popular neu- ral models can improve the ability of learning text representation by integrating knowledge, because they usually lack the ability to learn entities and their relationship in the text. However, the medical ∗Corresponding author. 1https://github.com/Gdls/HiCapsRKL Title: The clinical observation of amiodarone combined with metoprolol in the treatment of elderly patients with ventricular premature beats. Abstract: The study analyzed the clinical efficacy and safety of amiodarone combined with metoprolol in elderly patients with premature ventricular contractions. Ventricular premature beats are one of the most common arrhythmia in our country. ***. Contrast the occurrence of adverse reactions such as chest pain after taking the medicine.*** Keywords: amiodarone; metoprolol; elderly ventricular premature beats; control test; adverse reactions Medical Literature Knowledge <Metoprolol, Indications, Arrhythmia> <Metoprolol, Adverse reactions, Chest pain> <Amiodarone, Combination, Metoprolol> …… Figure 1: An example to show the undifferentiated knowledge in the medical literature. The literature is from the Chinese Medical Association and is translated from the Chinese version. ‘***’ represents the omitted texts in abstract. The entities are highlighted with italic, and the sentences with underline describe the relation- ship of two entities. literature usually contains multiple knowledge, and not all knowledge is beneficial to its subject. The integration of undifferentiated knowledge into the neural models may reduce the …

Content truncated to 500 words