Wenyi Shang is a third-year Ph.D. candidate of the School of Information Sciences at the University of Illinois, Urbana-Champaign. Working with his advisor, Professor Ted Underwood, his research focuses on digital humanities. Prior to joining the University of Illinois, he earned a bachelor’s degree in information management at Peking University, China. His current research investigates different types of humanities data, such as large-scale literary texts, historical documents, data in relational databases, and bibliographic metadata, in order to address humanities problems (mostly on history and literature). He has published in multiple digital humanities and information science journals and conferences:
Dr. Win-Bin Huang is an associate professor (tenured) in the Department of Information Management at Peking University, China. He also works as the director of undergraduate, master, and Ph.D. programs of Big Data Management and Application in the department and the director of Informationize and Information Management Research Center at Peking University. His areas of interest include data science, knowledge discovery, informetrics, and machine learning/deep learning. Before that, he had several years’ industrial experiences on IT and Solar System companies. He is a member of ACM and IEEE.
This is the source
In late 11th century, a reform carried out by Wang Anshi (1021–1086) brought about controversies and initiated a series of political struggles between factionalized reformers and anti-reformers. The origin and nature of these factionalized struggles have been discussed for a long time among scholars. In this paper, we discuss the issue based on the literary and political relationships among people in the era of the reform. First, two matrices are respectively constructed of the literary and political relations among these people based on the data collected from CBDB (China Biographical Database). Then a Poission-Gamma factorization model is adopted to obtain the key factors of the matrices, and the Louvain Modularity algorithm is used for community detection. The results show that people engaging in similar literary pursuits were more likely to share political interests and people belonging to the same literary groups were more likely to join in the same political groups, suggesting. Ensuing discussions illustrate that people’s differing academic views indeed played a shaping role in the formation and exacerbation of factionalized struggle, for which the mechanism unfolded herein of “literati politics” was highly responsible.
Illustrating a feature of “literati politics” which was prominent in 11th century China via a Poission-Gamma factorization model and the Louvain Modularity algorithm.
The political struggle among factions in the light of the reform in late 11th century China is considered as a decisive period in the Northern Song dynasty (960–1127). Beginning in 1069, Wang Anshi carried out the
[unceasingly impoverished and weakened] situation of the state. Thejipinjiruo
Because of the controversial nature of the reform, debates on it have never stopped.
Since the reform was a decisive political issue of that time, all contemporary officials
had to take their sides, and many of them had a fairly radical stance. For example, one
anti-reformer, Yang Shi (1053–1135) claimed that the origin of the disaster today was
because of Wang Anshi
The historical and cultural context of the reform should also be addressed. The Song
dynasty featured Tang-Song
transition
by Naito Konan
But this view is still not widely accepted, and the debate on the origin of the political struggle among factions in the light of the reform is not yet settled. Different scholarly views chiefly rise from different interpretations of existing historical sources, which are themselves not sufficient enough in many cases to serve as concrete pieces of evidence. By introducing digital methods and observing existing historical data in a more comprehensive way, however, it is possible to gain more solid arguments for the problem.
Thanks to the proliferation of digitized historical records, digital methods have been
adopted in many studies of Sinology. Specifically, a relational database for Sinology,
CBDB (China Biographical Database)
Following Cheng Yangzhi, but based on a totally different methodology, we try to provide a supplement to understand how close the links among scholars and politicians are by investigating the relationships among people who are active in late 11th century when the reform took place. Our study tries to verify the following hypotheses:
Here, the first two hypotheses can be directly tested with exploratory experiments.
Verifying them with quantitative evidence would strongly suggest the existence of the
feature of
We use quantitative methods to explore and texture literary and political affiliations among people. First, two correlation matrices constructed from the data collected from CBDB indicate their connections in literary and in political aspects, respectively. The features of the matrices, having been processed by dimensionality reduction, are then extracted as literary pursuits and political interests. Afterwards, Louvain Modularity is applied for community detection. Our experiment results in some interesting points, which can also be confirmed with primary historical sources.
We present a literature review on discussions of the origin and nature of the political struggle and the literati politics in part two. Our method is outlined in part three, including data collection, dimensionality reduction and community detection. The findings are shown in part four based on our results, and discussion in the context of the primary historical sources is given in part five. We offer a conclusion in part six.
There are four typical views towards the origin and nature of the factionalized struggle
in the light of the reform. The dominant view in the late 20th century Chinese academic
world is that the struggle was rooted in different class interests. Yang Rongguo is a
representative scholar holding the view Your majesty rules the nation with the scholar-officials, instead of the ordinary
people
The second view, chiefly held by Liang Qichao, imputes the political struggle to
emotionality
, and his main point is the members of each faction weren’t necessary to
combine with each other purposefully. However, they will raise disputes whenever an
issue occurs. If there is one person initiating an argument, there will be hundreds of
henchmen following him. In a word, they just acted out of emotionality
The third view is to attribute the problem to different regional ideas. It is proposed
by Hu Shi in his comments on Li Gou (1009–1059), a philosopher and a countryman of Wang
Anshi. Both Li Gou and Wang Anshi were from Jiangxi, and Wang’s thought was an
inheritance of Li
(the criticism of the anti-reformers to Wang Anshi) was actually based on different
views of the south and the north,
and that the southern literati are considered
frivolous while the northern literati are thought to be esteemed, thus Sima Guang
(1019–1086) and his followers were incompatible with Lü Huiqing and his followers
The last view, held by Cheng Yangzhi, is to ascribe the problem to different academic
opinions. He states that Sima Guang’s opposition to Wang Anshi’s reform was because of
his political belief and his basic principle, the
All of the views above were proposed in the first half of the 20th century, and many
later scholars stayed within the bounds they suggest. Luo Jiaxiang, who wrote a group of
papers on this in the 1980s and 1990s, defines the reformers as those who take the
responsibility of the nation
and the anti-reformers as the political group which is
both decadent and inert
they are both the representative of the whole class of landed
gentry
decadent and inert senior politicians
Recent scholars have proposed new perspectives. Li Xianchen thoroughly opposes the
traditional view about different class interests, arguing that The dispute between Wang
Anshi and Sima Guang was one within the reformers. Both of them advocated reformation,
and they were just different in specific methods of reform
Both
Finally, most of the scholars agree that the political struggle among factions in the
light of the reform made a profound impact on the Northern Song politics no matter what
they think the reason was. For example, Chen Zhenyang attributes the failure of the
reform to the struggle among factions. He states that [the anti-reformers] were glad of
the increment of the disadvantages of the reform and the nonfulfillment of its
advantages furtively,
and that Wang Anshi’s reform was in trouble because of it
[
A distinct feature of premodern Chinese government (which became an autocratic monarchy
governed by scholar-officials in the Ming and Qing dynasties), literati politics
played a crucial role in the cultural-political life of the Song Dynasty. For example,
in a case study on the Huizhou, in the period of 1200–1550, local literati’s spatial
imagination resembled a mode that every local place was a microcosm of
‘all-under-heaven’, and thus they practiced political ideas on their local governance
Tang-Song transition
by the academic world.
The idea of the Tang-Song transition
was brought up by Naitō Konan, who argues that
the fall of the Tang dynasty was the collapse of aristocratic government in China, and
the rise of monarchial autocracy and populism
local elite gentry families
replaced the
semi-hereditary professional bureaucracy
The influence of the literati on imperial politics was prevalent in the Song dynasty.
Hilde De Weerdt finds that the monarch, the court, the local officials, the
elite,
and is a broad generic reference to the group dominant in government, which also was
the paramount group in society.
The concept gradually transformed
from a warrior
caste…into a non-hereditary, ill-defined class of bureaucrats among whom litterateurs
were most highly esteemed
(421, item 5200).invoked the image of a turning point in the Family Instructions
Moreover, in the Northern Song, the development of the literati politics also
effectively carried cultural interactions among literati into the realm of politics. For
example, Amy McNair finds that many eminent scholar-officials active in the eleventh
century, including Fan Zhongyan (989–1052), Ouyang Xiu (1007–1072), Su Shi (1037–1101)
and many others, promoted the calligraphy style of Yan Zhenqing (709–784) because of
ideological considerations the centrality of
travel in the lives of the scholar-officials and the travelers’ representations of their
experiences
The commonness and impact of the literati’s practice of writing prefaces and postscripts
to each other is attested in the idea of
[writings are disseminated because of the prefaces], which is another manifestation of
the cultural-political relationships of the literati. The literati changed the function
and significance of the genres of prefaces and postscripts to gradually endow them with
robust political significance, especially when they wrote prefaces or postscripts for
political memorials
These phenomena also influenced the political struggle among factions. For
scholar-officials, their academic views were intimately related to their political
views, exerting a deep influence on the political struggle in late 11th century. Liang
Qichao noticed their relationship. In his analysis on
[the new school of Whatever
useful for molding the character of oneself or exerting an influence on politics are all
included in his academics
the opposition to
if Wang Anshi had not come into power or enacted a reform,
his academic ideas…would certainly have been inherited by the Confucianists working on
Confucianism principles for a long time
In conclusion, with literati politics
, a regime in which scholar-officials dominated
political life, culture and politics formed a dynamic and interactive relationship
during the Song dynasty. In the case of the large-scale political struggles among
factions after Wang Anshi’s reform, the interaction between literati and politicians, as
well as the political nature of various literary activities, can throw light on the
ensuing debates concerning their origin and nature. In the following sections, we will
therefore start by studying the literary and the political relationships among the
contemporaries during the reform and discuss the origin and nature of the political
struggle.
A framework including four data processing steps used in our study is shown in Figure 1.
In the first stage, data collection, a list of people active in the era of reform is
generated from CBDB, which is a relational database developed from primary historical
sources. Based on the list, the data about relationships among them are extracted, which
act as original data in the next stage. In the second stage, matrix construction, the
relationship data of literature and politics are processed and constructed into the
literary and political matrix respectively. The matrices are operated under the third
stage, dimensionality reduction, which mainly extracts the feature factors of the two
matrices, and the fourth stage, community detection, where communities are detected from
these matrices with Louvain Modularity. Both the results of the third and the fourth
stages are visualized and compared, while in the fourth stage, relation analysis is
adopted after comparison. Afterwards, the overall results are interpreted and discussed.
The details of the frameworks are discussed hereinafter
In the stage of data collection, we first generate a list of people active in the era of
reform. Since people recorded in CBDB are arranged chronologically from various
dynasties, we selected from the database only those who were active in the era of reform
from the second year of Xining (1069) to the last year of Yuanyou (1094), 1064 being the
year when Wang Anshi started the reform and 1094 the year before factional struggles
turned into thorough political persecutions. Furthermore, the people selected must be at
least 15 years old in 1094 since the age of 15, called
[an age of aspiring to learn] in Chinese culture
After generating the list, we continue to extract the relation data among the people in
the list generated from CBDB. In order to collect the literary relations among the
people selected, the concept of bibliographic coupling two people who have active literary relation to
a common third person
in our study. Here, active literary relation
means the person
is writing
(prefaces and postscripts, for instance) rather than being written for
.
Two people writing for a common third person implies that they share a mutual favor of
this person, suggesting a closer literary bond. However, in the case of two people being
written for by a common third person, there isn’t necessarily any such bond, since the
third person may do this simply out of the two other people’s great reputations.
The method of extracting data about political relations is similar to that for literary
relations, but data of direct interactions between two people, rather than their
relations with common third person, are collected. Besides, these interactions are
further divided into three types: explicit positive relations, implicit positive
relations and negative relations. Explicit positive relations include
After data collection, the data extracted are used for matrix construction. Two
correlation matrices — a literary matrix and a political matrix — are constructed, based
on the data of literary relations and political relations respectively. Assume that
there are nnactive
literary relation
with a common third person, the value ofji
Similarly, the political matrix is noted as nnij
where
In total, 5436 records of literary relations were collected, involving 144 people who act as the person to write for another person, and 2504 people who act as the person to be written for. 1453 records of political relations among the people were collected, with ninety-six people involved.
Next, both matrices are operated under dimensionality reduction, which discovers
compact representations of high-dimensional data
by generating a set of factors
theoretically and empirically more sound
in analyzing data with exact zero values
After running the model, twelve factors were obtained from each matrix. Similar to co-citation and bibliographic coupling analysis, where key factors are explained as an academic interest shared by a group of scholars, the historical meaning of the factors can be preliminarily explained as literary pursuits and political interests respectively. This interpretation is based on the following considerations: when two or more people share literary relationships with many common people, it is likely that they are driven by a common literary pursuit; likewise, when two or more people frequently offer political aids to each other, it is highly possible that they share a certain political interest. Furthermore, the mathematical meaning of factors that a single point could have a notable value in different factors matches with the fact that a person could have various literary pursuits and could be connected to different political groups through different political interests.
Therefore, the people sharing a common factor can be interpretated as engaging in a common literary pursuit and sharing a certain political interest respectively, despite the fact that the precise meaning of the key factors requires further elaboration. The first extracted factors explain higher variance, and therefore are chosen as the objects of evaluation. We first discuss the people with the highest value in each factor, and offer visualizations of the results afterwards. Finally, we compare the results of dimensionality reduction of the two matrices.
The last stage is community detection, which is also based on the two matrices. As a process of investigating the structure in a complex network, the method can aggregate and integrate all relation data among the people. Therefore, rather than adopting basic statistical methods, which separately analyze the relationships between each pair of people, community detection is capable of treating the entire social network in the reform era as a whole and analyzing it comprehensively. In this way, the results of community detection and dimensionality reduction, which suggest possible relationships among people regarding literary pursuits and political interests, can be complementary, providing a more comprehensive view of relationships among the people active in the era of the reform.
In community detection, for the purpose of eliminating the algorithmic bias caused by
negative values, since the minimum value in the political matrix is minus four, each
element is increased by a constant four so that all elements would be non-negative. Then
the matrices are converted into social networks, where the weight of edge between the
jiji
Similar to dimensionality reduction, the results of community detection are analyzed
separately and comparatively. When comparing the results of literary groups and
political groups, the research adopts methods of relation analysis, referring to the
main idea of association rule learning to evaluate the similarity between them.
Association rule learning is intended to identify strong rules discovered in databases
using measures of interestingness 00
And it defines the 00
According to the basic idea of association rule learning, association rules are required
to satisfy both a minimum support and a minimum confidence constraint at the same time
00
Twelve key factors of a matrix are produced after dimensionality reduction on the
literary matrix. Mean squared error (MSE) is adopted to measure the difference between
the estimator and what is estimated, the factors that are extracted from the original
matrix. Randomness or unaccountable information in the estimator is the major cause of
the difference, thus the value of the MSE is positively related to the error
The algorithm is run iteratively 400 times, and the MSE of dimensionality reduction of the literary matrix is 0.233. Since the first extracted factor is not interpretable as all people have the same value on it, we chose to evaluate the factors ranking from second to fourth. And for each factor, we picked the ten people with the highest values. The factor matrix is shown in Table 1.
Through the results of dimensional reduction of literary relations, people who have the
highest value in factors listed in Table 1 can be explained as engaged in common
literary interests
.
People with the highest values in factor
People with the highest values in factor
[four outstanding students of the
Cheng’s school] and Chen Guan is also deeply influenced by the two Chengs
(Cheng Hao
and Cheng Yi). Thus, the factors
The result of dimensionality reduction of literary matrix with factor
The political matrix is also reduced to twelve factors, which are explained as
political interests
. The algorithm is run iteratively 400 times and the MSE of
dimensionality reduction of the political matrix was 0.395. The first-extracted two
factors are selected and the ten people with the highest value are picked.
People with the highest values in factor
[Stele of the
was created in 1104, and allegedly
aimed at marking the officials who harm the politics in Yuanyou Era
(i.e. the
anti-reformers). However, it has a complex nature. Because of the monopolization of Cai
Jing, the prime minister who created the stele, the 309 officials listed in the stele
were actually far beyond merely anti-reformers. Rather, Cai Jing used the stele as a
tool to suppress his political opponents. Even one of the firmest reformers, Zhang Dun,
was listed in it because of his hostility to Cai Jing. Therefore, it shouldn’t be
considered as an objective tool to distinguish the anti-reformers, although it is the
only available historical source which stated to list the anti-reformers.
People with the highest values in factor
[accomplice of Wang Anshi] by Liang Tao (1034–1097)
The result of dimensionality reduction of the political matrix with factor
Synthesizing the results above, we find that the reformers are all mostly in Sect.
The correlation matrices are operated under Gephi, and a literary network and a political network are generated based on the data of the matrices respectively. After laying out and partitioning the networks, they are divided into several communities. The result of community detection of literary network is shown in Figure 4, with the size of the nodes representing the importance of the person:
There are four core communities detected in the network, each representing a group of literati. Other communities are dispersed in the edge of the network and consist of only one component, thus should be considered as unimportant. The findings are as follows:
by Liang TaoWang Anshi qindang
Of course, this categorization is reductive: people cannot be categorized into merely
one or another literary or political group. However, as found above, despite some
exceptions (such as the leader of reformers, Wang Anshi, is in the core group
The modularity of the political network is similar to that of the literary network. Since the matrix is high-density and most of the values in the matrix are around four (if two people don’t have associations in politics, the value of the element demonstrating the strength of their relation is four owing to our normalization during preprocessing), the degree range of the filter is set to be “greater than four” so that all of the edges shown connect two people that are positively related in politics. The result are as follows:
Figure 5 shows fifty-four people in seven core communities. In a manner similar to the literary network, other unimportant communities are dispersed in the edge of the network and consist of very few components (at most three). The constituents of the seven core communities all have positive political relations with at least one person in the network.
The communities, each representing a political group, are reasonably clear. Following
the labels shown in Figure 5, the analyses of Group
), and Liu Shu. Sima Guang is the most important person in the group and the two distinct members ofYuanyou dangji bei
Wang Anshi qindang
These findings demonstrate the reliability of the data in the political matrix, and the
network accurately reflects the political situation in the era of reform. An exception
can be found in group
Finally, we analyze the relations between literary groups and political groups referring to the main idea of association rule learning. The groups considered as related are shown in Table 3:
The results of relation analysis show twelve pairs of related political and literary
groups, and they imply a close relationship between these groups. For example, political
group
In general, the results confirm our original hypothesis that those who shared a literary group were more likely to share a political group.
In Figure 5, there is an outlier in the political network: Wang Anli is found to appear in the group of mild anti-reformers. This outlier is derived from the following historical record:
Su Shi was imprisoned in the jail of the Censorate. He was in extreme danger and no one dared to save him. Wang Anli said (to Emperor Shenzong) calmly, ‘since ancient times, magnanimous emperors do not convict others because of their sayings.…If Your Majesty sentenced him (Su Shi) strictly according to the law, I am afraid that Your Majesty will be criticized by later generations as intolerant to the talents.’…Thus Su Shi was only punished lightly.
This is far from the only record showing positive interactions between reformers and
anti-reformers. Actually, Su Shi, the leader of
Wang also listened to the advice of Su Zhe, who was the brother of Su Shi. When he
carried out the reform policies, he asked for the opinion of Su Zhe on the What you said was really reasonable, I should rethink it
carefully.
And he stopped talking about
Similar records are not only found in those concerning the amiable relations between the
reformers and the members of
Moreover, even the firmest anti-reformers tried to mediate with the reformers. In the
fifth year of Yuanyou (1090), Liu Zhi (1030–1098), the leader of mediation
We next investigated the difference between the reformers’ and the anti-reformers’ political views, in order to discover to what extent they really hold irreconcilable political views.
The anti-reformers who held the least radical views on the reform were still the members
of Nowadays, if we do not
cleanse the old, we will not be able to establish something eminent
talked about improving government
finance in his early years, but never talked about it after seeing the failure of
Su Shi was not the only one who urged reform. Although criticizing Su Shi, Zhu Xi
admitted that not only
Moreover, during the era of the reform, members of
In conclusion, before the era of the reform, most anti-reformers still approved of
reform to a certain degree. And during the era of the reform, the members of
Nevertheless, the situation worsened as time went on. The first to be persecuted were
the members of
In the Yuanfeng Era (1078–1085), the reformers were intolerant of him and wanted to kill him. In the Yuanyou Era, there was also dissension between him and the old gentlemen (i.e. the anti-reformers). He is not a man to fluctuate as situation changes.
As a result, he was relegated in the Xining Era (1068–1077) and the Yuanfeng Era when the reformers were in power, and was also forced to leave the capital in the Yuanyou Era when the anti-reformers were in power.
The members of In my opinion, the origin of the factions is because of political views.
Those who have the same view with me are considered to be righteous, while those who
have different views are doubted to be devious
In the Southern Song dynasty, when the arguments around the reform came to the
definitive conclusion that the reform had been a mistake, people’s judgement on
reformers and anti-reformers completely depended on their political views: the
anti-reformers received praises and honors, while the reformers, no matter who they
were, were severely criticized. In
In 5.1 and 5.2 we discussed positive interactions between reformers and anti-reformers
and the anti-reformers’ views of the reform. We concluded that, at first, there were no
irreconcilable conflicts between reformers and anti-reformers. But we found that
conditions of political civility deteriorated in the last decades of 12th century, and
the findings in part 4 offer a persuasive explanation for this: what we have called
throughout literati politics
. We conclude that those engaged in common literary
pursuits were more likely to share political goals, and that shared affiliation to
literary groups is likely to indicate shared affiliation to political groups. Both
conclusions suggest the basic constitution of literati politics, which in turn exerts a
deep influence on factionalized struggle.
The relatively mild view held by members of
Zhu Xi, a representative of the orthodox Confucianists, stated that Wang Anshi and Su
Shi regarded the Buddha and Laozi
the
academics of both of them were illicit.
If Su Shi had been the prime minister, and invited Qin Guan
(1049–1100), Huang Tingjian and other companies to assist him, things would have become
worse (than what was caused by Wang Anshi)
The similarities between the philosophical perspectives of the members of
The distinct features of literati politics in the Song dynasty led to the extremely
important social status of the literati. The actions of scholar-officials tended to be
value-rational
instead of instrumentally rational
literatus
and official
at the same time, their
academic views would strongly influence their political views, which brought about
political struggles. Even worse, the scholar-officials in the Song dynasty were
concerned with Confucianism, which itself pertained largely to value judgements.
Therefore, they criticized those who held different academic views, and eventually gave
moral judgements on them. This was the chief reason why normal political struggles would
turn into disastrous persecutions according to
Nevertheless, a paradox remains, namely that in the Southern Song dynasty, the dispute
between
With the help of digital methods, this article preliminarily confirms that it was the
literati’s different academic views that caused and exacerbated the political struggle,
and the regime of literati politics
was largely responsible for it. In part four we
have found a distinct feature of literati politics from the results of our data
analysis. And in part five, beginning with an outlier found in the automatically
generated social network, we went back to the primary historical sources, and found that
there were not irreconcilable conflicts between the reformers and anti-reformers at
first. Further, in order to explain the reason why the political struggle worsened, we
found that the development of literati politics in the Song dynasty offered a persuasive
cause.
In this article, the original data obtained from CBDB are entirely based on historical
records, and the reliability of the findings generated by digital methods that
distantly
aggregate historical sources are further validated by traditional methods
that closely
look through such sources. This is particularly necessary as both
dimensionality reduction and community detection algorithms adopted in this article
belong to the class of unsupervised learning. However, it should be noted that, without
access to the workings of CBDB, work in this field would be held back. Moreover, the
specific ways in which literati politics impacted political struggle still need
elaboration in further studies, and our methodology can also be adapted by researchers
in other fields such as exploring correlations between economic and political
perspectives.
With the above caveats in place, we can conclude that, as found in part four, although
literary relations and political relations among people have their own characteristics,
they have many clear connections during the era of the reform, making it a typical era
with highly-related culture and politics. Therefore, based on the suggestiveness of the
data and the confirmation offered by the primary historical records, this article
reinforces the view held by Cheng Yangzhi on the origin and nature of political struggle
surrounding the issue of reform, ascribing the problem to different academic opinions
This result may shed light on further studies on the origin of the political struggle.
This paper offers one way to bring to a close the debate on the origin and nature of the
struggle among political factions in the reform era. Of course, other factors such as
class interests, regionalism and simple emotionality still have weight, but literati
politics offers the most comprehensive probative value of any of the possible
explanations. Moreover, our paper also proves the explanatory force of the Tang-Song
transition
model
Niu Xi imputed the origin of the political struggle among The disputation had nothing to do with policies, nor with principles. It was
just a disputation based on regionalism, parochialism, companionship between teachers
and students, and academic styles
It is generally acknowledged that the relation between the scholars and politicians became increasingly close in medieval China, and the feature of literati politics became more and more significant through the dynasties. Consequently, China eventually transformed into an autocratic monarchy governed by scholar-officials in the Ming and Qing dynasties. The Song dynasty acted as a pivotal period of the transformation with its distinct feature of literati politics, under the rubric of which scholar-officials holding different academic views criticized the political opinions of others, such that even political persecutions could be entirely driven by moral judgements.