一种基于关联度分析的学术社会网络搜索算法研究

发布时间：2018-10-29 10:08

【摘要】：学术社会网络是通过学术活动构建起来的网络,学者组成了网络中的各个节点,学者之间的合著关系构成了网络中的边。随着学术研究越来越快的发展,学术社会网络的规模也逐渐增大。在规模如此巨大的学术社会网络中搜索需要的信息,属于比较前沿的研究方向。目前,已经有很多学者对学术社会网络搜索进行了研究,也已经取得了阶段性的进展。将这种学术搜索付诸实践,论文审稿人评选就是其中一个典型的应用。它是在专家搜索的基础之上考虑审稿人与被审稿人之间的社会关系,从而搜索出符合条件的审稿人。为了解决学术社会网络搜索问题,本文提出一种基于关联度分析的学术社会网络搜索算法。本研究课题提出的算法主要内容包括:首先,需要计算候选节点与查询节点之间的内容相似度,这里采用的是短文本相似度的计算方法。为了使得计算出来的内容相似度更加全面,也更加符合实际情况,本文提出一种基于邻居节点语义关联度的短文本相似度计算方法,能解决之前短文本相似度计算方法存在的不足;其次,需要计算候选节点与查询节点之间的结构相似度,它是利用节点之间的最短路径来表示的,由于本文研究的网络图属于无向无权图,因此可以采用广度优先遍历的方法计算最短路径;然后,计算候选节点的权威度。综合上面的三个因子,构造候选节点与查询节点之间的关联度模型。最后,采用随机游走搜索策略进行节点的搜索,为了使得搜索过程更加快速准确,提出基于最短路径的随机游走搜索策略。这样,经过上述过程每个节点都会有一个分值,根据分值的高低为候选节点排序,选择指定数目的节点返回给用户。本文使用C-DBLP的数据集对搜索算法进行性能测试。实验结果表明,基于关联度分析的学术社会网络搜索算法在各项性能指标上比其他的搜索算法都有所提升,与之前的理论推断相符合。
[Abstract]:Academic social network is a network constructed through academic activities. Scholars make up each node of the network, and the co-authorship between scholars constitutes the edge of the network. With the rapid development of academic research, the scale of academic social network is gradually increasing. Searching for the required information in such a large academic social network is a frontier research direction. At present, many scholars have carried on the research to the academic social network search, also has made the stage progress. To put this kind of academic search into practice, the selection of paper reviewers is one of the typical applications. It considers the social relationship between the reviewer and the reviewer on the basis of the expert search, so as to search the qualified reviewer. In order to solve the problem of academic social network search, this paper presents an academic social network search algorithm based on correlation analysis. The main contents of the algorithm are as follows: firstly, the content similarity between candidate nodes and query nodes should be calculated, and the method of calculating the similarity between candidate nodes and query nodes is adopted here. In order to make the content similarity calculated more comprehensive and more in line with the actual situation, this paper proposes a short text similarity calculation method based on neighbor node semantic correlation degree. It can solve the shortcomings of the previous similarity calculation method. Secondly, we need to calculate the structural similarity between candidate nodes and query nodes, which is represented by the shortest path between nodes, because the network graph studied in this paper belongs to the undirected unauthorized graph. Therefore, the shortest path can be calculated by using the breadth-first traversal method. Then, the authority of the candidate node is calculated. By synthesizing the above three factors, the correlation model between candidate node and query node is constructed. Finally, the random walk search strategy is used to search the nodes. In order to make the search process more rapid and accurate, a random walk search strategy based on the shortest path is proposed. In this way, each node will have a score after the above process, according to the value of the candidate node sort, select a specified number of nodes to return to the user. This paper uses the data set of C-DBLP to test the performance of the search algorithm. The experimental results show that the academic social network search algorithm based on correlation analysis has better performance than other search algorithms, which is consistent with the previous theoretical inference.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【参考文献】