Material performance evolution discovery based on entity
extraction and social circle theory⋆
Jinzhu Zhang1, Wenwen Sun1
1 Department of Information Management, School of Economics and Management, Nanjing University of Science
and Technology, Nanjing China
Abstract
Topic evolution analysis describes the emerge, develop, and extinct of topics in a field, which can
help researchers understand the history and current situation of the research field. However, the
material patent text has a certain domain specificity, and the general entity extraction models
cannot extract special entities effectively. Moreover, the belief that topics with high similarity have
evolution relationship contradicts the rule of “first the change, then the new topic”, which cannot
clearly present the dynamic changes and accumulation of topics. Therefore, we design a method to
extract the material performance entities accurately and construct dynamic evolution path for
material performance topics. Firstly, we propose a material entity extraction model BERT-BiLSTM-
CRF, which integrates syntactic dependency analysis and attention mechanism, realizing the
accurate extraction of material performance entities. Secondly, we design an algorithm for
identifying the evolution relationship between performance nodes based on ring boundaries, which
can mine the evolution relationship between performance nodes and existing topics, realizing the
dynamic accumulation and change of topics. Finally, we construct the dynamic evolution path of
material performance, exploring the complex associations of material performance. Experiments in
the field of metal materials confirm that the proposed method can effectively construct the dynamic
evolution path of material performance topics, which makes the evolution relationships between
topics more abundant and interpretable.
Keywords
Entity extraction, material performance evolution, patent entity relationship1
1. Introduction affect the material's microstructure. So, current
researches perspectives on the evolution of material
With the emergence of a large number of patents on performance are mainly divided into the three
materials manufacturing and materials innovation, it has perspectives: "performance evolution - microstructure" ,
become critical to explore the complex associations and "performance evolution - microstructure - manufacture
evolutionary trends of material performance. Such process" , and "performance evolution - manufacture
exploration can help researchers deepen their process". The specific relationships are shown in Figure
understanding of material performance and promotes 1. Among them, Perspectives I and II [2, 3, 4] involve
the invention of new materials [1]. microstructure, require high levels of expertise and
Through the review of existing studies, we learned that experimental equipment from researchers and readers ;
the material performance mainly depends on their Perspectives III [5] usually proceed in the form of
microstructure, while the manufacture process directly
Joint Workshop of the 5th Extraction and Evaluation of Knowledge
Entities from Scientific Documents and the 4th AI + Informetrics
(EEKE-AII2024), April 23~24, 2024, Changchun, China and Online
EMAIL: zhangjinzhu@njust.edu.cn ( Jinzhu Zhang ); Sun000216@16
3.com ( Wenwen Sun )
© Copyright 2024 for this paper by its authors. Use permitted
under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
72
controlling variables, such as varying the temperature of 2.1. Data sources and preprocessing
a certain process, then exploring the influence of on the
We considered that to do material evolution, if we
microstructure of the material and the evolution of the
collect data at random intervals, it may lead to a lack of
material performance, which to some extents restricts
completeness and accuracy in the final analysis of the
the comprehensive understanding and description of the
evolution results.So we takes the concept of Germany’s
material performance evolution.
III
“Industrie 4.0” as the background, and selects metal
affect material as an example, which is one of the key
foundational materials closely related to this concept.
I II
material
determine
material s
change
manufacture Then we use the Derwent Innovations Index database as
performance microstructure process
the patent data retrieval platform. The patent search
Figure 1: Different Perspectives in the Analysis of The expression is “TS=(‘Metal materials’ OR ‘Metallic
Evolution of Material Performance materials’ OR ‘Metal alloys’ OR ‘Metal compositions’ OR
In terms of evolution analysis methods, researchers ‘Metal-based materials’) AND WC=(‘Materials Science’)”,
usually use the following approaches:(1) topic with a time interval from 2011 to 2023, where 2011 is
identification [6]; (2) topic evolutionary analysis [7, 8]. the year when the concept of "Industry 4.0" was first
However, the former is difficult to reveal document introduced. Then, the top 10,000 relevant patent texts
semantics effectively, resulting in poor interpretability of were selected as the dataset. In addition, considering the
the topics [9] . The latter only reflects the relative number of patents in each period, we divide it into year
importance or attention of a topic at a specific time point batches for material performance evolution.
[10]. Moreover, most of the existing studies are based on
topic similarity, believing that there is an evolution 2.2. Method for extracting the material
relationship between two topics whose similarity above entities
a threshold [11, 12]. But the topic itself is dynamically
Under the background of the continuous
changing and accumulating, there should be “ first the
improvement of the manufacture process, the
change in material performance, then the new material
processing method of materials is constantly progressing.
performance topic ”, so the establishment of an
At the same time, the change in the manufacture process
evolution relationship based on the similarity between
of the material will bring about changes in the material
topics is biased.
performance [13].
Therefore, this paper takes the material
Therefore, we defined the performance entity and
performance as the research object. Particularly, we
manufacture process entity of metal material. In
propose a method for constructing the topic evolution
addition, we refer to the relationship shown in Fig. 1,
path by introducing the social circle theory, realizing the
establishing the causal relationship between the two for
dynamic accumulation of material performance topics
subsequent analysis. (Among them, the manufacture
and the construction of evolution path. Finally, we
process entity and causal relationship will be used in our
explore the complex associations in the evolution of
next step of exploring the reasons for performance
material performance on the basis of the dynamic
evolution, so it is rarely involved in this paper.)
evolution path.
Then, considering the content of material patents
contains a large number of technical terms, material
2. Data and method
components, we constructed an entity extraction model
This paper takes the material performance as the (BERT-BiLSTM-CRF) by combining syntactic dependency
research object. Firstly, we integrate syntactic analysis and attention mechanism. The combined use of
dependency analysis and attention mechanism to these methods provides a more accurate extraction of
construct the material entity extraction model (BERT- the material performance entities and manufacture
BiLSTM-CRF), we obtaining the performance nodes of process entities from patent contents, providing a basis
each material. And then divided the performance nodes for subsequent material analysis and research.
of all materials into time batches by year. After that, we
designed an algorithm based on the initial performance
topics, to realize the dynamic accumulation of material
performance topics and the construction of evolution
path.
73
2.3. Method for constructing dynamic 2.3.2. Definition of evolution types
evolution path for material performance This paper defines six evolution types based on
topics existing studies. Among them, the four types of develop,
In this part, we first define six evolution types, then evolve, emerge, and fuse are derived from four different
we designed an algorithm for identifying the evolution social relationships in the social circle theory, and the
relationship of performance nodes based on ring two types of extinct and split refer to the existing studies
boundaries. Finally, we present the detailed process of to ensure the diversity of evolution types. In addition,
the method for constructing dynamic evolutionary path this paper also improves the fuse and split types, by
of material performance topics. further refining the different contributions of each
theme in them, which helps to consider the dynamic
2.3.1. Social Circle Theory interactions between themes in more detail. See
Appendix B for details.
Social circle theory suggests that the social circle
formed around a person reflects the closeness of his or 2.3.3. Identifying the evolution relationship of
her social relationships. That is, a person's intimate social performance nodes based on ring boundaries
circle usually consists of relationships with a high degree
of relevance; then followed by the normal friends circle We refer to social circle theory and improve the
and the strangers circle. In addition, there may exist such model proposed by Zhang et al. [14], proposing
a part of people in the sea of people: they are algorithm for identifying the evolution relationship of
temporarily outside your normal friends circle, but there performance nodes based on ring boundaries, the
are certain similarities between each other, and they specific algorithm and its correspondence are shown in
may become your friends or even intimate friends in the Figure 3. Firstly, for the existing performance topics, the
future, so this paper defines them as potential friends. centroid of each topic is calculated, and the maximum
Therefore, centered on the individual, their affinity rank Euclidean distance between each topic’s patent and its
order is: intimate friends, normal friends, potential centroid is taken as the topic boundary. The topic
friends, strangers, and the position belongs to: within boundary is extended outward by a ratio less than 1 to
the intimate friends circle, outside the intimate friends obtain the outer ring boundary and shrunk inward by the
circle within the normal friends circle, outside strangers same ratio to obtain the inner ring boundary. After
circle within the normal friends circle, outside the several comparison tests, we finally set the ratio in this
strangers circle, specifically as shown in Figure 2. study to 0.2.
evolve
intimate inner
intimate friends circle boundary emerge
C
friends circle
D
Strangers develop
Normal
Friends normal topic centroid1 A
friends circle boundary
normal Potential
Friends
Intimate
Friends
Strangers friends circle outer B
fuse
Normal Strangers circle
boundary
Mutual
Friends Friends
centroid2
Potential topic
Intimate boundary
Friends Friends
Strangers circle
Mutual Figure 3: Algorithm for Identifying the Evolution
Friends
Relationship of Performance Nodes based on Ring
Boundaries (See Appendix C for Detailed Picture)
Subsequently, for material performance nodes in
subsequent batches, the Euclidean distance between
each performance node and the centroids of existing
Figure 2: Social Circle Theory (See Appendix A for performance topics is calculated separately to identity
Detailed Picture) the evolution relationship between them. The specific
We refer to this theory and combine it with existing rules are as follows:
research to define six evolution types. And propose an a) develop: inside the inner ring boundary
algorithm for identifying the evolution relationship of b) evolve: outside the inner ring boundary and
performance nodes, the details are shown in parts 2.3.2 inside the outer ring boundary
and 2.3.3 respectively. c) emerge: outside the outer ring boundary
d) fuse: boundary intersection
74
Then, hierarchical clustering is introduced to obtain 55 and 60. The result of the evolution path are shown in
the different types of performance topics in the batch, Figure 5, where yellow, red, and green represent the
and merge similar topics that exceed a threshold (we set years 2021, 2022, and 2023 respectively. (see Appendix
it to 0.8 in this paper). For a topic in the previous batch, E for the entire picture, where Cluster stands for topic).
if its number of topics obtained more than two in this
batch, the evolution type is considered as split.
Furthermore, in the construction of the evolutionary
path, if a topic has no evolution relationship with the
following topics, we consider it as extinct type.
2.3.4. Construction of the dynamic evolution
path for metal material performance topics
The construction of the dynamic evolution path of
metal material performance mainly includes the
following steps, which are shown in Figure 4. Firstly, after Figure 5: Evolution Path of Metal Material
the extraction of performance entities of each material, Performance from 2021 to 2023
we get the performance node of each material, and then, From the results of the evolution path above, it can
all material performance nodes are divided into time be observed that in 2022, Cluster1 developed from the
batches according to the year. Secondly, the K-Means Cluster1 in 2021. The contents of the two Clusters are as
algorithm is used to cluster the first batch of data to follows: [‘high excellent low corrosion’, ‘resistance good
obtain the initial performance topics. powder strength temperature’], [‘high excellent alloy
Subsequently, for performance nodes in subsequent low mechanical’, ‘strength corrosion resistance process
batches, the algorithm for identifying the evolution good’]. It is not difficult to find that both Clusters focus
relationship of performance nodes (see Section 3.3.3 for on improving the corrosion resistance and mechanical
the specific algorithmic process) is used to identify their performance of metal materials, which aligns with the
evolution relationships with each performance topic. practical application requirements of metal materials
Then, hierarchical clustering is introduced to obtain the [15].
performance topics of different evolution types in this In 2022, Cluster1 further developed and split into
batch, and merge similar topics that exceed a threshold. Cluster1 and Cluster2 in 2023. The contents of the three
Finally, incremental iterations are carried out in the Clusters are as follows: [‘high excellent low corrosion’,
above manner to obtain the material performance topics ‘resistance good powder strength temperature’], [‘good
at different year batches, thereby achieving the dynamic surface heat’, ‘resistance layer wear low’], and
construction of the material performance evolution path. [‘resistance good base layer wear’, ‘layer wear surface
(see Appendix D for the entire picture, where Cluster low heat’]. As can be seen, the 2023 Cluster1 maintained
stands for topic) the original corrosion resistance and wear resistance
Cluster1
One time slice Cluster2
performance of Cluster1 in 2022, and further improved
the surface heat performance of metals. This may be
Metal Material Data flow Cluster3
Patent Abstract Text
achieved through improvements in material technology
and alloy additions, thus enhancing the performance in
evolve
emerge split Cluster1 evolve
practical applications.
fuse
develop
develop
extinct Cluster2
4. Conclusion
Figure 4: Process of Constructing The Evolution Path of This paper proposes an algorithm for identifying the
Metal Material Performance evolution relationship of performance nodes based on
ring boundaries, which can not only realize the dynamic
3. Result accumulation and construction of metal material
In the construction of the evolution path of metal performance evolution path, but enrich and improve the
material performance, we use examples from the years topic evolutionary analysis method. Currently, we are
2021-2023 to obtain the results. Specifically, the number combining the manufacture process entities of each
of performance clusters in 2021, 2022, and 2023 are 66, material to further analyze the causes of the evolution of
75
material performance in depth, and to better [7] Parlina A, Ramli K, & Murfi H. Theme Mapping and
understand the evolution trends and the changing Bibliometrics Analysis of One Decade of Big Data
patterns of material performance. Research in the Scopus Database. Information,
2020,11(2): 69. doi: 10.3390/info11020069.
Acknowledgements [8] Zibiao Li, & Li Zhang. Evolution of Patented
Technology Topics of Steel Materials Based on LDA
This work is supported by the National Natural
Model. SCIENCE AND TECHNOLOGY
Science Foundation of China (No. 72374103, 71974095)
MANAGEMENT RESEARCH, 2020, 40(24): 175-183.
and the Postgraduate Research & Practice Innovation
doi: 10.3969/j.issn.1000-7695.2020.24.023.
Program of Jiangsu Province (No. KYCX23_0632).
[9] Behrouzi S, Sarmoor Z S, Hajsadeghi K, & Kavousi K.
Predicting scientific research trends based on link
References prediction in keyword networks. Journal of
[1] Kaushal Jha, Suman Neogy, Santosh Kumar, R N Informetrics, 2020, 14(4): 101079. doi:
Singh, & G K Dey. Correlation between 10.1016/j.joi.2020.101079.
microstructure and mechanical properties in the [10] Wu Z, Xie P, Zhang J, Zhan B, & He Q. Tracing the
age-hardenable Cu-Cr-Zr alloy. Journal of Nuclear Trends of General Construction and Demolition
Materials: Materials Aspects of Fission and Fusion, Waste Research Using LDA Modeling Combined
2021, 546, 152775. doi: With Topic Intensity. Frontiers in Public Health,
10.1016/j.jnucmat.2020.152775. 2022, 10(3), 22-26. doi:
[2] Enyu Guo, Guohua Fan, & Tongmin Wang. 10.3389/fpubh.2022.899705.
Microstructure Evolution Mechanism of Metallic [11] Jinxia Liu, Qianqian Hou, Jing Du, Fuhou Chai, & Li
Materials: Progress on In Situ Studies Using Zhang. Hot Topics Evolution Research of Emerging
Synchrotron Radiation Source. Failure Analysis and Field Under the Sub Topics and Vocabulary Related
Prevention, 2021, 16(01): 1-14+91. doi: Perspective. Journal of Intelligence.
10.3969/j.issn.1673-6214.2021.01.001. 2023,42(3):123-129. doi:10.3969/j.issn.1002-
[3] ZhiGuo Liu, Ri You, & Jintao Wu. Study on 1965.2023.03.017.
relationship between dam concrete material [12] Reza T H S, Sadegh A, & Soroush T. An embedding
performance evolution and its pore structure. approach for analyzing the evolution of research
WATER RESOURCES AND HYDROPOWER topics with a case study on computer science
ENGINEERING, 2016, 47(11): 25-28+35. subdomains. Scientometrics, 2023,128(3): 11-16.
[4] Yinzhi Li, Gang Liu, Xinzhe Chen, Zhenhua Su, doi: 10.1007/s11192-023-04642-4.
Lianwu Yan, & Yingbiao Peng. Effects of Sintering [13] Qing-Miao Hu, & Rui Yang. The endless search for
Temperature and Holding Time on the better alloys. Science, 2022, 378: 26-27.
Microstructure and Properties of Ti(C,N)-Based doi:10.1126/science.ade5503
Cermets[J]. Packaging Journal, 2023, 15(03): 37-45. [14] Zhang Y, Zhang G, Zhu D, & Liu J. Scientific
[5] Yang Liu, Baoliang Liu, Shi Kai, Xiaoyuan Han, Yi Xia, evolutionary pathways: Identifying and visualizing
& Jianzhao Shang. Effects of different zirconia- relationships for scientific topics. Journal of the
based raw materials on properties of MgO-Al-C Association for Information Science and
materials. REFRACTORIES, 2023, 57(01): 59-64. doi: Technology , 2017, 68(8): 1925-1939. doi:
10.3969/j.issn.1001-1935.2023.01.013. 10.1002/asi.23814.
[6] Lee J W, & Han D H. Data Analysis of Psychological [15] Y.B. Lei, Z.B. Wang, B. Zhang, Z.P. Luo, J. Lu, & K. Lu.
Approaches to Soccer Research: Using LDA Topic Enhanced mechanical properties and corrosion
Modeling. Behavioral Sciences, 2023,13(10), 43-47. resistance of 316L stainless steel by pre-forming a
doi: 10.3390/bs13100787. gradient nanostructured surface layer and
annealing. Acta Materialia. 2021, 208, 116773. doi:
10.1016/j.actamat.2021.116773.
76
A. Appendices
intimate
friends circle
normal
Strangers friends circle
Normal
Friends
Potential
Intimate
Friends Friends
Strangers circle
Mutual
Friends
Figure 2: Social Circle Theory
B. Appendices
Table 1 Six patterns of topic evolution types (where Cluster stands for topic)
social Evolution Expression and illustration Explanation
relationships types
develop
Cluster1 Cluster2 Cluster2 only develops
intimate friends develop
from Cluster1
evolve
Cluster1 Cluster2 Cluster2 only evolves from
normal friends evolve
Cluster1
evolve develop
Cluster1 Cluster2 Cluster3
Cluster4 is treated as an
strangers emerge
emerge Cluster which has
Cluster4
no evolutionary
relationship with previous
Clusters.
Cluster1
cluster3 comes from
mutual friends fuse Cluster3
cluster1 and cluster2
Cluster2 together.
77
Cluster2
Cluster1 is divided two or
/ split Cluster1
Cluster3
more Clusters
evolve develop
Cluster1 Cluster2 Cluster4
Cluster3 is extinct which
/ extinct has no evolution
Cluster3
relationship with the
following Clusters
C. Appendices
evolve
intimate inner
friends circle boundary emerge
C
D
Strangers develop
Normal
Friends normal topic centroid1 A
friends circle boundary
Potential
Intimate
Friends Friends fuse
outer B
Strangers circle
boundary
Mutual
Friends
centroid2
topic
boundary
Figure 3: Algorithm for Identifying the Evolution Relationship of Performance Nodes based on Ring Boundaries
D. Appendices
Cluster1
One time slice Cluster2
Metal Material Data flow Cluster3
Patent Abstract Text
evolve
emerge split Cluster1 evolve
fuse
develop
extinct Cluster2
develop
Figure 4: Process of Constructing The Evolution Path of Metal Material Performance
78
E. Appendices
Figure 5: Evolution Path of Metal Material Performance from 2021 to 2023
79