
Juanzi Li(李涓子)
Professor
Sign in to view more

Papers416 papers
Patents
Patentee: 清华大学唐杰,张静,荆榆,刘德兵,李涓子
本发明公开了一种快速计算图节点相似度的方法,包括步骤:对于给定的图,进行路径采样并构建从节点到路径的倒排索引;对于所述图中的每个节点,根据所述倒排索引得到每个节点的共现路径节点,计算每个节点和所述共现路径节点之间的近邻相似度;对于所述图中的每个节点,选择top?k个近邻相似点的相似值作为每个节点的特征向量;对于所述图中的每个节点的特征向量构建kd?tree索引;根据所述kd?tree索引,计算所述图中每个节点向量的top?k个结构相似点。本发明具有如下优点:能够统一且快速地计算近邻和结构两种相似度。
Publication Patent Date: 20160713Application Date: 20160223Patent type: 发明专利Legal Status: 实质审查Publication Patent Number: CN105760503A
本发明涉及一种异构网络中的推荐方法以及基于该方法的推荐系统。异构网络中的推荐方法使用一种统一的模型来模拟多种不同类型的对象以及他们之间可能存在的复杂关系,并借助该模型更好地完成推荐。首先建立社会网络库(其中的对象及其关系均为异构);然后利用重要程度计算装置,对社会网络对象进行全局重要度评估;接着利用相关程度计算装置,对社会网络对象进行相关度评估;也可以再利用浏览历史信息计算装置,得到当前用户的浏览信息;最后利用上述三个基本装置,在社会关系网中,进行给定子图、推荐子图的操作。该推荐方法切实有效,克服了现有技术中只能推荐某一种类型对象的缺陷。
Publication Patent Date: 20081210Application Date: 20080704Patent type: 发明专利Legal Status: 有效专利Publication Patent Number: CN101321190
Individual relation finding method based on path grade at large scale community network
A personal relationship discovery method based on path score in the large-scale social network, belongs to social network search technique field in network, it is characterized in that define path score based on weight first, then find out the shortest path between every two people, then find out all the multiplying path that the distance between the appointed two people is not more than the shortest path, return all the path according to the sequence of path score to users. The invention can be applied to network with more than 1 million panel points to quickly find out the relationship between people or used in the relationship found of researchers.
Publication Patent Date: 20080326Application Date: 20071109Patent type: 发明专利Legal Status: GrantedPublication Patent Number: CN101149756 A