所属 コンピュータサイエンス学部 コンピュータサイエンス学科 職種 専任講師
|著者・共著者||伏見 卓恭，佐藤 哲司|
|概要||A large amount of documents are posted on the Web
from moment to moment such as news articles, blog articles, web pages, academic literature.
There are strong and weak relationships between related and similar documents.
The relationships between strongly relevant documents clearly exhibit like
citations of scientific literature, trackbacks of blog posts,
hyperlinks of Wikipedia articles and web pages,
but in the case of news articles,
connections with related documents are often not clearly indicated.
As a simplest method,
there is a method of calculating similarity between news articles
and constructing a similarity network by linking between similar documents,
but it is difficult to consider the time axis.
Therefore, in this paper, we propose a topic forest construction method
consisting of multiple time-evolving tree structures
based on semantic cohesiveness and temporal cohesion of documents.
By visualizing this topic forest,
it is considered that an effective access order to the document can be presented.
Experimental evaluations using real data show that
the topic forest has semantic and temporal cohesiveness,
which helps us to improve accessibility to the documents.