Several Technology Of Scientific Literature Knowledge Mining

The electronic scientific literature grows in explosive with the rapid development of network technology. The massive scientific literature provides us with tremendous reservoir of knowledge. And these literatures can not only greatly save researchers' times of searching the literature and put themself into the research work quickly, but also better achieve the knowledge sharing, if those literatures be used reasonable. So the mining for scientific literature has become a primary subject.What the main purpose of literature mining is extracting the implicit, valuable knowledge and rules from huge and chaotic literature data. Our mining work of scientific literature carried out on two problems. Fist, how to locate the articles of someone interested quickly and the articles' quality must also be high. Second, how to obtain the major research topics in the given field form the literatures. Meanwhile, let users know the research matters that other researchers focusing on in the field and the core research organizations for a given field.This paper presents a solution to fast locate articles of somebody interested, which is the cluster-topic pair mining. However, considering the large number of related literature will be returned for a search keyword. And more import, some articles are not worth reading for researchers. So, we present a linear model to assessment the reading value of every paper, which weights are adjusted by a method based on expert feedback. Then we can pick the high value of papers for text clustering. In addition, we extract keywords to describe the every cluster. After three steps of last described, we will get many pairs of papers cluster and corresponding describe words. Thus, the user can decide to read or not to read these papers. And it is convenient to retrieval literature for users.Second, In order to allow the user having an intuitive understanding and grasping the whole research topic for a given field, this paper carried out second data mining, which is hot analysis about research issues and institutions. The hot analysis for research topic mainly using a cluster method based on the relationship of keyword co-occurrence, and we use a visual way to display distribution graph of the field in a certain period. And we also try to discovery the core research institutions for a certain field, and during the process we assess each institute considering the number of researchers, the volume of published articles and the amount of the citation comprehensively.In summary, the contribution of our work is mainly reflected in two aspects. First in case of no much information loss, we select the papers according to the access model of read value presented in this paper, during the cluster-topic mining. The cluster performance can improve after paper selection. What the second contribution is we present the hot evaluation method for research institutes, and we can draw the hot trends for different institute in a given field.