基于频度演化的领域知识关联关系涌现

Emergence of Correlation in Domain Know ledge Based on Frequency Evolution

领域知识的生长演化问题一直是图书情报学界重点关注的主题。以网络科学思维探索知识生长过程中的关联关系涌现问题,能够对知识关联的生长模式与机制进行揭示。本研究提取知识关联关系累计44万余对,关联频度87万余次,共划分为11个时间窗口。在对时间序列领域知识网络结构属性初步判识的基础上,对关联频度分布进行时间序列分析。并从领域知识生长过程中的关联频度数量、关联频度占比、邻近窗口状态等方面,对知识关联关系及其频度进行跟踪与分析。研究结果表明,知识关联关系生长过程中,关联关系频度的分布符合幂律分布,且在领域知识发展的成熟期表现得更好。知识关联关系的生长过程具有频度层面的"富者更富"的属性,且主要遵循"择优强化"机制。尽管研究所使用的基于社会化标注系统的知识网络尚不足以囊括所有类型的知识网络,但是基于频度演化的知识关联关系涌现模式与机制,有助于促进知识网络、知识生长等领域的研究工作,对于社交网络、传播网络、交通网络等研究不无裨益。

The growth and evolution of domain knowledge have always been the focus of Library and Information Science. Exploring the emergence of correlations in the process of knowledge growth with network science thinking can reveal the growth patterns and mechanisms of knowledge correlations. The present research extracts a total of over 440 000 pairs of knowledge correlations, 870 000 times of correlation frequency,which are divided into 11 time windows. The small-world and scale-free properties of time series domain knowledge network topology are determined by the short average path length, the high clustering coefficient and the power-law distribution of degree. On this basis, the knowledge correlations and frequencies in the process of domain knowledge growth are tracked and analyzed from the aspects of the number of correlation frequencies, the proportion of correlation frequencies, the status of neighboring windows, etc.Results have shown that in the process of knowledge correlation growth, the frequency distribution of correlations is in accordance with power-law. Through the time series analysis of the frequency of knowledge correlations,it is found that the frequency distribution of knowledge correlations in the same network can better fit the power-law than the degree distribution of nodes, without the phenomenon of ＂top-heavy distribution＂ usually found in the degree distribution of nodes, and the power-law distribution of the frequency of knowledge correlations performs better in the latter part of the time series. This phenomenon shows that although the knowledge correlations determine the topology of knowledge networks, only a few correlations have extremely high frequency values,while most of them have only a small number of frequency values, and the process of frequency of knowledge correlations is that of frequency emergence. On the other hand, the growth process of knowledge correlations has the property of ＂the rich gets richer＂ in frequency,and mainly follows the mechanism of ＂preferential reinforcement＂.Statistics show that as the frequency of knowledge correlations grows and accumulates, the few frequency ＂rich＂（high-frequency correlation） occupy more and more frequency ＂wealth＂ than most frequency ＂poor＂（low-frequency correlation）. The gap between the ＂rich＂ and the ＂poor＂ in frequency of knowledge correlations becomes more and more obvious in the second half of the timeline as the domain knowledge grows and develops. The status of the neighboring windows shows that most of the knowledge correlations in the frequency ＂extremely rich＂ status often have the ＂extremely rich＂ status in the previous time window.Correlations with more frequency ＂wealth＂ will attract additional frequencies with a higher probability in the process of growth and development of domain knowledge, reflecting the frequency growth mechanism of ＂preferential reinforcement＂. The repeated superimposition of the micro-rules of ＂preferential reinforcement＂ on the time series have created the phenomena of ＂the rich get richer＂ at the frequency level in the emergence of correlation. The study has also found the ＂bursts reinforcement＂ phenomenon that the correlations in frequency ＂extremely poor＂ status suddenly jump into the ＂extremely rich＂ status. The main reason for this phenomenon is that academia has produced significant findings or inventions,which provide a possibility for identifying significant academic achievements based on the frequency of knowledge correlation.Although the knowledge networks based on social tagging system used in this study are not comprehensive to cover all types of knowledge networks, the emergence patterns and mechanisms of knowledge correlation based on frequency evolution help to promote research in knowledge networks and knowledge growth, and they also benefit studies on social networks, communication networks, transport networks, etc.