=Paper=
{{Paper
|id=Vol-2456/paper73
|storemode=property
|title=A User Interaction Demonstration For Exploring Medical Knowledge Graph
|pdfUrl=https://ceur-ws.org/Vol-2456/paper73.pdf
|volume=Vol-2456
|authors=Jin Xiao,Lianfen Miao,Chunyu Li,Ruimin Lu,Bo Chen,Yuan Ni
|dblpUrl=https://dblp.org/rec/conf/semweb/XiaoMLLCN19
}}
==A User Interaction Demonstration For Exploring Medical Knowledge Graph==
https://ceur-ws.org/Vol-2456/paper73.pdf
A User Interaction Demonstration For Exploring
Medical Knowledge Graph
Jin Xiao1 , Lianfen Miao1 , Chunyu Li1 , Ruimin Lu1 , Bo Chen1 , and Yuan Ni1
PingAn Technology, Shanghai, China
Abstract. As the importance of the knowledge graph increasing in
many fields, finding an efficient way to display and interact with the
data is a huge challenge, especially in the medical field. We here re-
vealed an initial proof-of-concept demonstration, a web application. In
this demonstration, we will show the user interface to explore or get
information from the domain-specific real-world data. The demonstra-
tion exhibits the flexibility of displaying complex medical data, helps
users to understand and explore their data. We’ll have fun with a use
case of the demonstration. For more information, please visit our web-
site(http://121.12.85.245:1316/#/ ) or watch the video on YouTube(
https://www.youtube.com/watch?v=E4qw-_q5etY).
1 Introduction
A knowledge graph represents entities and their relations. In order to help users
or domain experts make intelligent decisions and support artificial intelligence
systems for medical applications, such as information retrieval, Q&A and so
on, we build a huge but high-quality medical knowledge graph in Chinese. Our
knowledge graph covers 2.7 million medical terms, including 1 million core terms
such as diseases, symptoms, drugs, body structure, examination etc. It also in-
cludes various medical relationship triples such as disease-symptoms, disease-
examination, disease-indications. An entity, such as heart failure(心力衰竭), can
be associated with many other entities especially in medical field. One example
is that heart failure − has symptom → cyanosis, as same as
triple(heartf ailure, hassymptom, cyanosis)
Based on the technique shown in Fig.1, our medical knowledge graph was con-
structed by six subsequent steps. We is constructed a Knowledge Graph in the
medical field.
1. Data Collection. Firstly, We obtain a lot of unstructured, semi-structured,
structured data from various sources, such as OMAHA1 , UMLS2 , ICD-103
,some medical websites and drug instructions.
1
https://www.omaha.org.cn/
2
https://www.nlm.nih.gov/research/umls/
3
https://www.who.int/classifications/icd/icdonlineversions/en/
Copyright © 2019 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
2 Jin Xiao, Lianfen Miao, Chunyu Li, Ruimin Lu, Bo Chen, and Yuan Ni
Knowledge Graph Application
Knowledge Graph
Entity Linking and Resolution
Knowledge Extraction
Data Processing
Data Collection
Fig. 1: The Architecture of Our Knowledge Graph
2. Data Processing. Cleaning data is the main task in this step.
3. Knowledge Extraction. Information Extraction from the data was han-
dled by last step, include named entity recognition and relation extraction.
4. Entity Linking and Resolution. Based on medical data reviewed by au-
thoritative medical experts, we integrate data from various sources,by ap-
plied for entity linking and resolution technology.
5. Knowledge Graph Construction. We store the graph data into the Neo4j
database.
6. Knowledge Graph Application. This knowledge graph empowers to many
applications in our product. This demonstration shows novel and efficient
user interaction of medical data.
There is a lot of work focused on how to visualize knowledge graph [1,2,3].
However, a visualization tool of a knowledge graph which is universal, effective
and concise rarely appears. Our demonstration is a tool what aim to make it
easier to explore and visualize knowledge graph data. The demonstration system
is composed by two components:
– api backend, which contains a graph database, supported by Neo4j 1 . and
a RESTful API used for providing data to frontend interface.
– frontend or interface, a web interface implemented by JavaScript and
HTML2 .
2 Demonstration of Use Cases
For constructing and using the knowledge graph, we used a dashboard for sum-
marizing the data, as showed in Fig.2.
1
https://neo4j.com/
2
Chrome browser is recommended
A User Interaction Demonstration For Exploring Medical Knowledge Graph 3
Fig. 2: The dashboard of demonstration
One the second tab, a hierarchical disease tree can be found(Fig.3), which
can be folded and unfolded by click.
Fig. 3: Hierarchical disease tree
When we click on the nodes 心血管系统疾病 → 心脏病 → 心力衰竭 in
order on the tree, a main view will be showed on the right of page to represent
the corresponding data where the center is clicked node. Furthermore, disease,
symptom and drug entities associated with it can switch to the center by clicking.
The encyclopedia about the disease which would be showed when the center
node being hovered. The content page contains semi-structured information such
as overview, diagnosis, treatment, follow-up and common complications.
The main innovation in our demonstration is the dynamic interaction graph
displayed on the right column. It consists of many rings, with lots of small dots
on it, which can be clicked when the category is disease, drug or symptom(Fig.4).
Different rings represent different node categories, category of the ring can be
seen in the legend under the graph. The English mapping can be found on Table
1. When clicking the rings, it will be highlighted and the related node will appear
4 Jin Xiao, Lianfen Miao, Chunyu Li, Ruimin Lu, Bo Chen, and Yuan Ni
around the center node. The bottom right corner is a track which can go back to
the node that was clicked in the previous steps. For more features, please visit
our website.
Fig. 4: Interaction snapshots of the demonstration
Table 1: The English mapping of Chinese keyword
Chinese English Example
疾病 disease 心力衰竭 • heart failure
症状 symptom 紫绀 • cyanosis
检查 test 放射性核素成像 • radionuclide imaging
发病部位 bodystruce where disease be found 心脏 • heart
手术 surgery 气管内插管 • endotracheal intubation
禁忌药品 constrained drugs 氨茶碱片 • aminophylline tablets
适应症药品 indication drugs 氯丙酸片 • chloropropionate tablets
易感人群 susceptible population 老年人 • elderly group
下位疾病 sub-disease of a disease 慢性心力衰竭 • chronic heart failure
3 Conclusions
Only first two disease tree can be clicked in this demonstration. However, the
approach to explore and visualize the graph data is novel and effective. We
proposed the interface way, showed in the demonstration, would help doctors,
domain experts or the public, who are exploring and searching information from
a knowledge graph. Meanwhile helping them get access to information and make
informed decisions more easily.
References
1. Antoniazzi, F., Viola, F.: RDF Graph Visualization Tools: A Survey. Conference of
Open Innovation Association, FRUCT 2018-November, 27–38 (2018)
2. Kharlamov, E., Giacomelli, L., Sherkhonov, E., Grau, B.C., Kostylev, E.V., Hor-
rocks, I.: Ranking, aggregation, and reachability in faceted search with SemFacet
1963 (2017)
3. Liebig, T., Vialard, V., Opitz, M.: Connecting the dots in million-nodes knowledge
graphs with SemSpect 1963 (2017)