Two Roads Diverged: A Semantic Network Analysis of Guanxi on Twitter

Guanxi, roughly translated as"social connection", is a term commonly used in the Chinese language. In this research, we employed a linguistic approach to explore popular discourses on Guanxi. Although sharing the same Confucian roots, Chinese communities inside and outside Mainland China have undergone different historical trajectories. Hence, we took a comparative approach to examine guanxi in Mainland China and in Taiwan, Hong Kong, and Macau (TW-HK-M). Comparing guanxi discourses in two Chinese societies aims at revealing the divergence of guanxi culture. The data for this research were collected on Twitter over a three-week period by searching tweets containing guanxi written in Simplified Chinese characters and in Traditional Chinese characters. After building, visualising, and conducting community detection on both semantic networks, two guanxi discourses were then compared in terms of their major concept sub-communities. This research aims at addressing two questions: Has the meaning of guanxi transformed in contemporary Chinese societies? And how do different socio-economic configurations affect the practice of guanxi? Results suggest that guanxi in interpersonal relationships has adapted to a new family structure in both Chinese societies. In addition, the practice of guanxi in business varies in Mainland China and in TW-HK-M. Furthermore, an extended domain was identified where guanxi is used in a macro-level discussion of state relations. Network representations of the guanxi discourses enabled reification of the concept and shed lights on the understanding of social connections and social orders in contemporary China.


Introduction
Guanxi is one of the key concepts scholars often use to describe Chinese societal features.It is the Romanization (pinyin) spelling of the Chinese character "关系" (in simplified Chinese) or "關係 "(in traditional Chinese).It can be translated as "social connections" or "personal relationships".However, in Chinese language, the word entails much richer meanings than social connections, and is bound to notions such as obligation and loyalty to ascribed social groups.As specified by the Oxford Dictionary of English (2010), guanxi is "the relationship that facilitates business and other dealings".This definition emphasises the functional role of guanxi in Chinese interpersonal relationships.
Guanxi relations sometimes refer to an exclusive and yet politically influential social clique that shares similar guanxi ties.One recent example of the guanxi network in politics comes from China's continuing struggle against corruption and conspiracy.A group of government officials were indicted for corruption earlier this year.According to reports on Chinese state media, this network was established among tongxiang (同乡), people from the same hometown, forming a "protective umbrella" to cover up bribery and embezzlement, leading to a "systematic corruption" (Yi, 2014).
Today, Confucian culture still exists in Chinese societies around the world, including Mainland China, Taiwan, Hong Kong, Macau, and overseas Chinese communities (Jensen, 1997;Barmé, 2000).
Guanxi culture, the social norms extended from traditional Confucian teaching, still exists in contemporary Chinese societies in the Mainland and in the greater Chinese regions, including Taiwan, Hong Kong, and Macau.However, these societies have gradually separated into distinct societies since the 1950s (Lin and Ho, 2009).
In Mainland China, Chinese traditional ethics and moral principles were first criticized and revolutionized with the establishment of People's Republic of China (PRC) in 1949 and the Cultural Revolution from 1960s to 1970s.The ideological revolution was reflected in the Party's avocation of simplified Chinese (jiantizi 简体字) in formal printing and writing.However, the post-Mao era has seen the decline of Marxism and the revival of traditional values.culture (Bell, 2010).The political rhetoric and the social orders that were valued in Confucianism, such as loyalty, family obligations, harmony, and concerns for others, have been once again emphasised in official and mainstream ideology.
Meanwhile, the greater Chinese communities outside Mainland China (for example, Taiwan, Hong Kong and Macau) demonstrate distinct socio-economic and cultural features due to having had different historical trajectories.The democratization of Taiwan took place in the 1980s and 1990s, which encouraged the public to participate in the political system and diversified the ideological system in Taiwan.As former colonies of Britain and Portugal, Hong Kong and Macau had been exposed to foreign cultures and ethics throughout their colonial history.Yet, Chinese culture and traditions still have an impact on the ethical and moral rules in Hong Kong and Macau (Wong, 1986).
Taiwan, Hong Kong, and Macau did not adopt simplified Chinese; thus, traditional Chinese still dominates both the informal and formal writing systems.As Chinese societies adopt modern values, questions have been raised within academia on whether or not Chinese traditional culture still has profound influence on Chinese contemporary interpersonal relationships, to what extend it has adapted to social changes that come with Chinese modernization, and how it has diverged in Chinese communities that have demonstrated different socio-economic backgrounds.

Conceptualising Guanxi
Early scholars studying guanxi culture traced its roots back to Chinese social structure.Chinese sociologist, Fei Xiaotong summarized the Chinese organizational principle into a "different mode of association" (chaxugeju, 差序格局) (Fei, 1992: 71), which resembles ripples spreading from the centre of a body of water when a stone has been thrown into it.Social connections in China are nonequivalent and rely strongly on family ethics and kinship positions.Western social structures, in his opinion, have equivalent social connections and are constituted by autonomous groups and memberships.Fei's observation of Chinese social structure echoes the words of Max Weber.Weber noticed that behind that Chinese rational ethics were the principles of moral laws, which constitute a "complex of useful and particular traits" (Weber, 1964: 235).Chinese characteristic social patterning could also be attributed to the ethical system in Confucian teachings, in which the fundamental traits in personal relationships are summarised into five basic ethics (lun 伦) (Liang, 1987).The similar patterning of family to non-family relationships also appear in Chinese five lun, which constitutes the basic social order within Chinese society (King, 1985).
Recent scholars have considered guanxi culture from a micro and interpersonal perspective.
Empirical researches taking this perspective proved the importance of kinship system in Chinese guanxi culture.Jacobs (1979: 237) defined guanxi as "particularistic ties", which consists of ascriptive ties, such as family members, or people from the same native place.Family and extended family ties are also the foundation and starting point of establishing other types of guanxi connections (Cohen, 1970;Baker, 1979).They sometime bridge between family ties with non-ascriptive ties and forms a sense of intimacy beyond the realm of immediate family.Jankowiak (2008) found the phenomenon where individuals address unfamiliar friends as family members such as sisters or uncles, in order to establish an intimate and useful connection.
As a word frequently used to describe social connections, guanxi is often compared with social network in the Western academia.Some scholars considered guanxi different from social network (King, 1985;1991), while empirical research on guanxi suggested that guanxi practices could not be fully interpreted using the framework of social network, and thus guanxi has certain features that make it essentially Chinese (Bian, 1994;Bian and Ang, 1997;Burt, 1992;Burt, 2002;Xiao and Tsui, 2007).
Another group of scholars disagreed with the uniqueness and Chineseness of guanxi (Lin, 2001;Wellman, 2001;Wellman, Chen and Dong, 2002).They tried to incorporated guanxi culture within a larger conceptual framework of social network.Nevertheless, they still acknowledge the cultural influence of guanxi in Chinese societies.

Functional Practices of guanxi in Business and Politics
Studies conducted in Mainland China highlighted guanxi as a source of informal political power besides formal bureaucratic power, and in some cases even "defuse and subvert the elaborate regulations and restrictions" (Yang, 1994: 320).The administration "governed by human relationships" (Pye, 1992: 29) hampered Chinese government credibility in the Mainland, and has still exist in Chinese contemporary political system (Gold, 1986;Bian, 2002;Tsai, 2007).
Market economies in both Mainland China and in TW-HK-M have prospered since the mid-1990s.Researchers observed a pragmatic practice of business guanxi in both Chinese societies.Studies on Chinese family companies located in both Mainland and other Chinese communities have found that family businesses often demonstrated nepotism and paternalism characteristics (Cohen, 1970;Wong, 1985).It is sometimes used as a business strategy to provide informal support and protections, and is utilized to raise venture capital (Hamilton, 1998;Wank, 1999;Xin and Pearce, 1996;Luo, 1997;Hsing, 1998).
However, mainland Chinese Government plays a relatively strong role in the market economy.
Study showed that business guanxi in the Mainland has been intentionally manipulated to establish connections with government officials in order to gain "back door access" in a state-dominated economy (Wank, 1999).
Turning to the 21st century, Chinese society in Mainland has moved towards a new model that demonstrates a combination of political socialism and economic capitalism.TW-HK-M, meanwhile, shares closer ties with the world economy.Under such circumstances, has guanxi, or social connections in China, co-evolved with other societal changes?Some scholars see guanxi as the cultural consequences of the ancient Chinese social structure.
Guanxi culture is thought to be challenged by two forces: one from political rationalization and one from the influence of market economy.
Political rationalisation and bureaucracy is believed to be an important factor accounting for the contemporary transformation of guanxi culture.Potter (2002: 183) argued that traditional Chinese guanxi will work as "a complement to rather than substitute for formal institutions" in the face of China's preceding legal reforms.Guthrie regarded guanxi as an "institutionally defined system" (Guthrie, 1998: 255) that could be shaped by the evolving political regime.
Chinese economic reform and the increasing influence of the market mechanisms are believed to be another driving force of the transformation of guanxi.For example, Yang speculated that abusive gift economy between government and market might gradually diminish with the market gradually replacing the state in resource distribution (Yang, 1994).Other researchers also documented the decline of cadre privilege (Walder, 1995) and the weakening of guanxi between officials and entrepreneurs (Ma and Cheng, 2010) with institutional changes in market-oriented reforms.

Our Contribution
As reviewed above, extensive research has contributed to the conceptualization of guanxi, and has examined the practice and transformation of guanxi culture.However, there remain two gaps in the literature: First, existing empirical research was mainly conducted during the early stages of Chinese reform in the 1990s, this was when China was still a largely socialist economy.As Chinese societies are undergoing dramatic social, economic and political changes, it is important to refine our understanding of guanxi in topics such as interpersonal relationships, informal politics and business network in contemporary Chinese society.Second, in many of the works on guanxi, the concept is often traced back to its Confucian root.However, these societies have followed divergent historical paths and developed different political and economic systems.Thus, it is important to be aware of the differences between Confucian societies inside and outside of Mainland China.
Based on the gaps identified in the literature, this research explores the following two questions: Unlike previous studies of guanxi, this research takes a unique approach to examining contemporary guanxi culture.Data come from everyday discourses on social media.Text data of guanxi discourses in this research were collected from Twitter, which is the only popular microblogging site that has users from both Mainland and the greater Chinese communities, providing an ideal public arena for a comparative analysis of two Chinese popular discourses.This type of evidence of how users talk about guanxi, rather than how they respond to surveys or what can be observed by scholars, reveals a much more diversified interpretation of guanxi than inspecting it within a particular social activity or contextualising it within a specific community.As Kipnis argued (1996), analysing language surrounding guanxi can explore the richness of guanxi culture without over-generalising the concept beyond its meanings.

Data Collection
Data in this research were collected from Twitter's Search API 1 .The keywords in the search queries were simplified Chinese guanxi character ("关系") and traditional Chinese guanxi character ("關係").Keyword search queries for recent tweets were sent every day over a period of three weeks from May 12 th to June 1 st , 2015.After combining the results and deleting duplicate tweets 2 , there were 18,833 distinct tweets containing simplified Chinese guanxi and 9,572 distinct tweets containing traditional Chinese guanxi.

User Location
This research used the different written forms of guanxi as proxies for two guanxi cultures.
To justify this argument about written Chinese as indicative of Chinese societies in different regions, the two groups of guanxi tweets were compared by user locations.Location information (where available) was extracted from the metadata of all tweets, and was used for the comparison of the geographic distribution of users tweeting in simplified Chinese and in Traditional Chinese.In this research, user timezone is chosen as the main indicator of locations (see Appendix A for more details on the geographic information about the tweets).
1 The returned result type was set to be "recent" tweets.According to Twitter's official documentation, returned results are likely to be incomplete.See: https://dev.twitter.com/rest/public/search 2 Each tweet has a unique id ("id_str" in every status object), which was used as filter to delete repeated tweets in original dataset.
The number and proportion of tweets from each region are given in Table 1.31.5% of tweets containing the Simplified Chinese character guanxi were sent by users located in Mainland China, but the percentage of users located in Mainland China among Traditional Chinese tweets was only 17.91%.
A large percentage of users tweeting the Traditional Chinese guanxi were located in Taiwan, Hong Kong, and Macau.Therefore, the user language, either Simplified Chinese or Traditional Chinese, overlaps with the regions in which the users are located.Noticing that 2.24% of Simplified Chinese tweets were sent by users from Singapore and Malaysia, these tweets were deleted to enable comparison between Mainland China and TW-HK-M China.The final dataset contained 11,417 tweets about Simplified Chinese guanxi and 9,572 tweets about Traditional Chinese guanxi.

Building Corpora
The two corpora in this research were organized and cleaned in three steps: punctuation filtering, word segmentation, Part-of-Speech tagging and filtering.In the first step, raw texts in Simplified Chinese and in Traditional Chinese were aggregated separately by extracting tweet text from the dataset, and were processed to filter punctuations, emoji and other symbols.Then a Python module for Chinese word segmentation and part-of-speech tagging, Jieba3 , was utilised to segment words in each sentence in both corpora. 4,5After tokenising, the corpora consist only space-spliced words.Not all words are necessary for understanding the context of sentences (Jurafsky and Martin, 2014).By denoting all words in both corpora with their according part-of-speech, we only keep words that are semantically meaningful, which includes nouns, verbs, adjectives and adverbs.

Word Co-occurrence
In understanding word senses, one group of linguists suggested approaching word meaning via its relationship with other words in the same sentence, taking a structuralism perspective (Firth, 1957;Saussure, 2011;Evans, 2005).To structuralise language in concrete forms, semantic network was proposed by applied linguists in the field of computer science (Quillian, 1963;Collins and Quillian, 1969).
As with other types of networks, semantic networks consist of vertexes and edges.Vertexes, or nodes, are concepts; edges can either be constructed by existing paradigmatic knowledge of linguistic relationships such as thesaurus (Fellbaum, 1998;Kozima and Furugori, 1993), or by the totality of structural relationships, such as co-occurrence (Freeman and Barnett, 1994;Danowski, 1993;Doerfel, 1998).
In this research, we used word co-occurrence to construct the semantic network.As defined by Freeman and Barnett (1994), the links in a semantic network are the co-presence frequency of a pair of concepts in a given slicing window of the text.The slicing window in this research was set to be three words6 , and WORDij7 was used to count the co-occurrence of word pairs and frequency of words.

Measuring Associations
The association of each word pair and can be measured by their co-occurrence frequency.
However, words in natural language are distributed following the power law (Jurafsky and Martin, 2014) (see Appendix B for information on word frequency distribution of corpora in this research).
Highly associated word pairs denoted by co-occurrence frequency, therefore, are likely to co-locate due to chances rather than semantically meaningful connections.For this reason, it is important to use a different measurement, one that can measure the likelihood that the target word pair ( cooccurred by chance irrespective of the frequency of the words.One such measurement using the ratio of observed occurrence and the expected probability of occurrence if two words appeared independently was proposed by Church and Hanks (1989), and is called Pointwise Mutual Information ( ): ( Where and are words' counts in the corpus, normalized by , the size of the corpus.measures the likelihood that and would co-occur in the same corpus if they were independent.is the joint probability of and observed to be co-located in a window of consecutive words. Using as a measurement of words association takes into account the scenarios, in which two words only locate within the slicing window by chance.However, it has two drawbacks as a cooccurrence statistic: First, the denominator could be extremely small for very rare words, making low frequency word combination scores disproportionately high, which would interfere with detecting more meaningful and commonly used word pairs.Second, there is a lack of an upper bound for PMI value, making it practically difficult to define a high PMI score.To solve these two problems, Bouma (2009) devised a method to normalise PMI, which controls the influence of rare words and defines an upper boundary.The normalised measure, nPMI, is defined as: (2) In this research, the scores of all the word pairs were used as the weight of the edge between the two nodes in the semantic network.

Visualising Semantic Network
After calculating the scores for all the word pairs in the Simplified Chinese and Traditional Chinese corpora, edges (weighted by score) and nodes (distinct words) can be visualised in two semantic networks: one in Simplified Chinese and the other in Traditional Chinese.
Both networks were further filtered to exclude rare words and edges with lower values.
First, we filtered out words with very small frequencies (see Appendix B); words that appeared three times or less in the Simplified Chinese and only once in the Traditional Chinese corpus.
Next, when choosing a threshold for significant word pairs, we used a percentile threshold for determining the most relevant word associations 8 .The major consideration in choosing a threshold is not whether the value is statistically significant, but whether it is practically important for The semantic network of guanxi in Simplified Chinese and in Traditional Chinese were displayed separately, each node represents a word from guanxi corpora.Node size indicating the node 8 Another method to identify significant edges is to shuffle the nodes (words), shuffle the network while remain the node degrees, and rewire nodes randomly with each other.By calculating t-value, a p-value can be found using Student's t-distribution.If p-value is less than the conventional threshold for statistical significance, 0.05, then the null hypothesis (two words are independent) is rejected and the alternative hypothesis (they are associated semantically) is accepted (see Newman, Watts and Strogatz, 2002).
strength, which is the sum of weights attached to edges belonging to a node (Barrat et al., 2004).Edges are undirected and weighted.Two semantic networks were visualised using network analysis tool: Gephi (Bastian, Heymann and Jacomy, 2009), displayed by applying Force Atlas 2 algorithm (Jacomy, Heymann, Venturini, and Bastian, 2014).Word pairs connected by edges with higher association values ( scores) are attracted proportionally closer than others.
Semantic networks contain a large amount of concepts and concept relations.To categorise concepts, community detections were applied on both guanxi semantic networks using Modularity partition algorithm in Gephi (Newman, 2006).After partitioning, the Simplified Chinese semantic network of guanxi was divided into 43 communities, with Modularity score equals to 0.431; the Traditional semantic network was partitioned into 24 communities, with a Modularity score of 0.3399 .
Themes of each sub-community were summarized by the central concepts, which were defined as the top 5 percent of nodes ranked by their node strength in descending order (see Appendix D for details on the justification of choosing the 5 percent threshold).Figure 3

Results
By comparing the concept sub-communities of the two guanxi semantic networks, we found that although the two discourse networks share similar themes, including interpersonal relationships, business-political networks, and state relations.However, they demonstrated different interpretations within these topics.

Guanxi in Interpersonal Relationships
As explained previously, guanxi consists of social norms and traditions relating to interpersonal connections.Through examining the Simplified and Traditional guanxi semantic networks, it can be seen that they both have one or several sub-groups concerning interpersonal relations.However, guanxi in the domain of interpersonal relationships have demonstrated different focuses in the two Chinese societies.

Mainland: Intimacy and Traditions
There are two concept communities in Simplified Chinese semantic network related to interpersonal relationships.The first one ranks third largest sub-community (see Figure E1 in Appendix E), highlighting intimate guanxi; the second one is the fourth largest sub-community (see Figure E2 in Appendix E) in the Simplified Chinese network, with a focus on Chinese traditions.
Familism in traditional Chinese culture consists of members within an extended family.
Particularly in rural China, the boundary of a family could be as broad as a lineage or a clan (King, 1991).However, as shown in Figure E1, the highest ranked concept community in family guanxi in Simplified Chinese is more concerned with the nuclear family than it is with lineage.Concepts related to romance are emphasised by relationships such as "Marriage relations" (夫妻关系), "Lovers" (情人) and roles such as "Wife" (妻子), and "Girlfriend" (女友).The extended family retains its influence on individual through interfering with marriage.For instance, remotely or closely related "Aunts" ("七大 姑八大姨" ) are common matchmakers who are actively involved in seeking and assessing an ideal match for the "Boy" ("男孩") or "Girl" ("女孩").
Concepts of guanxi as a tradition or cultural heritage belong to the fourth largest concept subcommunity in the Simplified Chinese semantic network (displayed in Figure E2).In this community, feelings and social consequences of guanxi tradition were discussed.Originating from "lun" in Confucianism, the guanxi tradition is characterized by "Patriarchy" (" 家长制 "), highlighting the hierarchical relationship between "Old generation" ("老人") and "Children" ("孩子") within the "household" ("家里").In addition, the traditions are connected through attributive ties, especially by "Consanguinity" ("血缘关系"), which is the sharing of bloodlines.
There is, however, a bittersweet emotion regarding traditional guanxi that is embedded in Chinese family morals.Centred on the deeply rooted guanxi traditions, there is nostalgia for a harmonious household and concern about overall social equality.On one hand, nostalgia for the guanxi tradition is linked to harmonious family relationships and cohesive family structure.For example, emotions like "Happiness" (" 幸福") and a sense of "Inheritance" (" 传承") are all inherent characteristics of the guanxi tradition.On the other hand, after Mao's anti-tradition social movement from 1966 to 1976, the "Cultural Revolution" ("文革") had challenged traditional values in urban and rural China tremendously, which is also reflected in this concept community via concepts such as "Revolution" ("革命").With the Chinese modernization process, there are also negative attitudes toward traditional guanxi norms.One criticism of guanxi tradition is the over-emphasis on one's family "background" ("背景").If used illegitimately to obtain private profits, guanxi might pose a threat to social "justice" ("公平").It might also jeopardize the legal rights and equal opportunities of other individuals from less privileged families.

Taiwan, Hong Kong, and Macau: New Wine in Old Bottles
Concepts regarding interpersonal relationships form the largest concept communities in the Traditional Chinese semantic network.This ranking order suggests discourses on guanxi are more often associated with social connections in TW-HK-M than they are in Mainland.The two Chinese communities also have varying emphasis on interpersonal relationships.As analysed in the Simplified Chinese network, intimate guanxi and guanxi heritage are heated topics in the domain of relationships.
In the Traditional Chinese guanxi network, however, central concepts are more general descriptions of social networks, without emphasizing intimate guanxi or debating on the traditions of Chinese family morals (see Figure E3 in Appendix E).
Chinese users in TW-HK-M show a form of interpersonal guanxi that is closer to the modern concept of social connections in many Western societies.traditional guanxi relations and the traditional moral requirements for the family are both missing form this sub-community.

Guanxi in Two Market Economies
As reviewed earlier, much research on the Mainland China and TW-HK-M markets has noticed the pervasiveness of guanxi practice in Chinese companies.Networks in this research also prove that guanxi in economic activities is an important topic in both societies, but business guanxi functions differently in the two contexts.

Mainland China: Guanxi and Rent-seeking
In one concept community of the Simplified semantic network, guanxi is frequently associated with rent-seeking behaviors, through manipulating guanxi ties with government officials.Concepts in this domain cover both the major players in rent-seeking processes and the corresponding social consequences of such practices.This concept sub-community on business guanxi in Simplified Chinese is displayed in Figure E4.

Taiwan, Hong Kong, and Macau: Business Network and Regulated Guanxi
In traditional guanxi semantic network, topics on business guanxi locate in separated community with political guanxi.Such a separation is different what is observed in simplified guanxi semantic network of Mainland users.
The second largest concept community in the Traditional Chinese semantic network consists of concepts related to business guanxi (see Figure E5).Unlike Simplified Chinese business guanxi, this concept community tends to focus on guanxi in economic activities as a business strategy, which separates the market from government officials."Enterprises" ("企業") and "Companies" ("公司") are major players in the business guanxi network.Guanxi in their business "Strategy" ("戰略") is considered to be access to personal "Communications" ("溝通") and as "Guanxi Capitals" ("人脈") for "Opportunities" ("機會") and "Investment"("投資").Compared to enterprises in Mainland, who have to invest in establishing guanxi resources with government officials; guanxi in TW-HK-M is a form of human resource and an informal business strategy brought by and maintained through personal ties between business partners.
Concepts of political guanxi are located in a different concept community in the Traditional Chinese semantic network, ranking as the fourth largest concept community (see Figure E6).In this concept community, political guanxi is confined to the legislative, executive and administrative divisions.Individuals who have political power are mentioned in a same political network, consisting of "Civil servant" ("公務員"), "Committee member" ("委員") and "Member of Parliament" ("議員").
Therefore, business guanxi in TW-HK-M is linked more closely to networks among personnel and less with governments.Political guanxi also raises alarms in these societies, but is restrained by written constitutions and formal laws.

Guanxi in Power Relations: An Extended Context
Most scholarly writings on guanxi considered the concept from micro-and meso-levels, focusing on the role of guanxi in networking activities between individuals or organisations.However, community-clustering on both guanxi semantic networks captured a new context of using guanxi, which is in power dynamics between states.In Simplified Chinese, most of the concepts of guanxi between states concerning diplomatic taxonomies and country names.Interestingly enough, it also demonstrates many diplomatic strategies borrowed from Chinese interpersonal social norms.For example, sentiments such as "Family" (家庭), "Trust" (信赖), and "Responsibility" (责任) are used in the domain of international relations.
This finding of state relations in both discourses has two implications: First, it suggests that Chinese guanxi culture is also reflected in the state's foreign policies.This expands the realm of guanxi study to the state level, and to Chinese diplomatic relations with other countries.As highlighted by the personalized sentiments in this concept community, the promotion of Confucian values plays an important role in Chinese diplomatic strategy.Second, the fact that over 16% of all concepts in Simplified Chinese guanxi semantics and 11% in Traditional Chinese guanxi semantics reside within this state relation community also indicates that in both Mainland China and in TW-HK-M China, Twitter is used regularly as news following and sharing services.This is also proved by the appearance of newspaper and agency names such as "People's Daily" (人民网), "VOA" ( 美国之音) in the two communities.

Discussion and Conclusion
This research uses guanxi discourses from social media to explore the transformation of guanxi culture in contemporary popular discourses.Table 2 provides an overview of the concept communities analysed in this research.The largest four concept communities in each semantic network cover three major topics concerning interpersonal relationships, business guanxi, and state relations.

Transformation of Family Values
We found that in terms of interpersonal relationships, both Mainland and TW-HK-M China have demonstrated an emphasis concerning the nuclear family.However, concept communities on interpersonal relationships show that the significance of traditional lineage families and their associated morals are declining in both the Mainland and TW-HK-M.
Despite the transformation of the family structure reflected in guanxi discourses, in the Simplified Chinese guanxi semantic network, special attention is still paid on the traditional values and morals in Mainland China.In the discussion of traditions, many concepts reflected negative attitudes towards the traditional system of hierarchy within the family and towards the over-emphasis of family background.
This widespread doubt about traditional family guanxi within the Simplified Chinese semantic network will be likely to result in a clash with official ideology.The transformation of the value system in official discourses aims at addressing the moral vacuum in contemporary Chinese society after the fall of Marxism ethics in face of the market economy (Bell, 2010).However, we can observe from the semantics of the guanxi tradition online that there is a mismatch between the popular discourses and the official ideology.This might lead to a sense of uncertainty and confusion about the ongoing transformation of ideology or doctrines in the Mainland.

Guanxi in Two Markets
Our research also showed in the Simplified Chinese guanxi semantic network, the concept guanxi bridges industries and the government, forming cliques in which members benefit each other through the exchange of political and economic resources.Such a phenomenon is also often the target of criticisms in the Simplified Chinese discourses.Meanwhile, business guanxi in the Traditional Chinese semantic network is separated from the political guanxi concept community.This clear division of politics and business ensures a fair environment for the market.
Whether or not political guanxi is intertwined with the business networks, and how political leverage is constrained to avoid corruption and rent-seeking behaviours mirror institutional differences in two Chinese societies.
The close-knit connection between political power and economic entities in the Mainland can be attributed to the lack of constraints over bureaucrats in Chinese state capitalism.Since the 1980s, the Chinese Communist Party has promoted a market economy featuring strong government control.With the market growth and capital accumulation, the lack of effective structural constraints on officials leads to rampant corruptions and rent-seeking behaviours.The lack of efforts for political reform might lead to the continuing spread of corruption and rising dissatisfaction concerning the party's legitimacy.
Eventually, this will result in criticism of and distrust in the government (Fukuyama, 2014).By contrast, in TW-HK-M, a globalized market and relatively effective constraints on bureaucratic power (Jones, 1994) help to prevent officials from using guanxi to obtain illegal profits.Guanxi connections in these market economies are more difficult to be manipulated by individuals.
A regulated market economy requires clear laws formalizing the types of guanxi practices that are acceptable and those that are forbidden.This is not only important for the Chinese market economy, but also significant for Chinese political legitimacy in the long run.Business guanxi in Taiwan, Hong Kong, and Macau demonstrates the possible variance of guanxi culture in a regulated market economy.
For leaders in the Mainland, these economies might provide solutions for tackling rampant corruptions.

Limitations
Whilst semantic networks generated from popular discourse could reflect the understanding of guanxi among Chinese Twitter users, there are still some limitations of the study.

RQ 1 :
Against the background of political rationalisation and economic reform, has the interpretation of guanxi changed in contemporary popular discourses?RQ 2: When compared Mainland China to TW-HK-M, to what extent does guanxi adapt to different economic, social and political environments?
interpretations.Based on qualitative comparison of different percentile thresholds, we applied a threshold of the 30 th percentile of sorted scores, which was 0.1955 for the Simplified Chinese corpus and 0.2546 for the Traditional Chinese corpus (See Appendix C for the justifications of choosing the percentile threshold).Two semantic networks consisting of edges with weights ( scores) exceeding the threshold values could now be constructed based on the filtered scores.After removing rare words and filtering unimportant word pairs, there were 20,710 words and 134,072 edges in the Simplified Chinese semantic network and 12,137 words and 119,461 edges in the Traditional Chinese semantic network.
and Figure 4 shows the clustered networks of Simplified Chinese guanxi and Traditional Chinese guanxi.Figure 2 summarizes the methodology of this research using a flow chart.

Figure 1 Figure 2
Figure 1 Methodology and Data Cleaning

Figure
Figure E7 and Figure E8 display the largest sub-community of the Simplified Chinese guanxi semantic network and the third largest sub-community of the Traditional Chinese guanxi semantic network; both communities contain concepts describing relationships between states.

First,
Twitter is unfamiliar to most of Internet users in Mainland China due to the government block (the Great Firewall); in particular, it has not been used by people who live in rural China or who do not have sufficient technological skills to bypass the Internet block.This makes Mainland Twitter users a relatively special group of Chinese Internet users and also lacks representativeness of the general Chinese population.However, there is an advantage in using discourses of Mainland Twitter users to study guanxi.This group tends to be very sensitive to Chinese politics and is less constrained by government censorship, which allows for the observation of critical opinions concerning Chinese political guanxi.Second, as the relationship between language and mind is complex, popular discourses on guanxi could not contribute more details regarding the dynamic process of practicing guanxi in offline settings.Third, concepts represented by segmented words, only provide a limited window into understanding what people think about or believe in, and thus certain context of the concepts is lost when processing language.To conclude, this research revisited guanxi culture in different Chinese societies by analyzing its linguistic representations in popular discourses.The concept guanxi bridges individuals and societies in the Chinese context.Reifications of the concept guanxi had not only provided understanding of guanxi in contemporary Chinese societies, but had also served as an interpretive device for studying the political and economic systems in which guanxi cultures are embedded.

Figure A 1
Figure A 1 Distribution of Simplified Chinese Tweets' Locations by Region, based on Tweet Country (in Percentage)

Figure
Figure E2 Simplified Chinese Concept Community Concerning Guanxi Traditions, fourth Largest Concept Community within the Simplified Chinese Guanxi Semantic Network (N=667, 3.21% of all nodes).Only central concepts were translated into English.

Figure
Figure E3 Traditional Chinese Concept Community Concerning Interpersonal Relationships.Largest Concept Community within the Traditional Chinese Guanxi Semantic Network (N=3224, 26.61% of all nodes).Only central concepts were translated into English

Figure
Figure E6 Traditional Chinese Concept Community Concerning Politics.Fourth Largest Concept Community within the Simplified Chinese Guanxi Semantic Network (N=1313, 10.81% of all nodes).Only central concepts were translated into English.

Figure
Figure E7 Simplified Chinese Concept Community Concerning State Relations.Largest Concept Community within Simplified Chinese Guanxi Semantic Network (N=3528, 16.97% of all nodes).Only central concepts were translated into English.

Figure
Figure E8 Traditional Chinese Concept Community Concerning State Relations.Third Largest Concept Community within the Simplified Chinese Guanxi Semantic Network (N=1358, 11.18% of all nodes).Only central concepts were translated into English.