全局视角下的网络社区多元知识关联挖掘

Holistic Perspective Multi-knowledge Relations Mining in Network Community

[目的/意义]网络社区中存在多种知识单元,知识单元间又有错综复杂的关系。有必要在保留知识单元全局信息的前提下,统一、简洁地开展多元知识关联挖掘。[方法/过程]提出网络社区多元知识关联挖掘的实现方案。首先,将网络社区中3种典型知识单元(用户、文本、词语)及其在知识交流中多种关系抽取为超网络;其次,利用网络表示学习算法将超网络中节点表示为统一特征空间下的低维稠密向量;最后,基于节点的向量开展多元知识关联计算。[结果/结论]以丁香园心血管论坛为例开展实验,验证方案的有效性。该方案既保留知识单元的全部信息,知识关联的挖掘又在统一低维特征下开展,且最终所得的知识关联满足网络社区知识组织场景多样性的要求。

[Purpose/significance]There are many knowledge units in the network community,among which there are intricate relationships.It is necessary to carry out multiple knowledge relations mining uniformly and succinctly on the premise of retaining all the relations of knowledge units.[Method/process]This paper puts forward the solution of multi-knowledge relations mining in network community.Firstly,3 typical knowledge units(users,texts and words)in the network community and their multiple relations in the knowledge communication were extracted into a supernetwork.Secondly,the network representation learning algorithm was used to uniformly represent the nodes in the supernetwork as low-dimensional dense vectors.Finally,multiple knowledge relations calculation was carried out based on nodal vector.[Result/conclusion]The effectiveness of the scheme was verified by taking cardiovascular BBS in dingxiang garden as an example.This scheme not only retains all the information of the knowledge unit,but also carries out the mining of the knowledge relation under the unified low-dimensional characteristics,and finally the knowledge relation meets the requirements of the diversity of the knowledge organization scene in the network community.