Automatic Keyword Annotation System Using Newspapers

抄録

Nowadays, an increasingly large amount of information exists on the web. Therefore, a method is needed that enables us to find necessary information quickly because this is becoming increasingly difficult for users. To solve this problem, information retrieval systems like Google and recommendation systems like that on Amazon are used. In this paper, we focus on information retrieval systems. These retrieval systems require index terms, which affect the precision of retrieval. Two methods generally decide index terms. One is analyzing a text using natural language processing and deciding index terms using varying amounts of statistics. The other is someone choosing document keywords as index terms. However, the latter method requires too much time and effort and becomes more impractical as information grows. Therefore, we propose the Nikkei annotator system, which is based on the model of the human brain and learns patterns of past keyword annotation and automatically outputs keywords that users prefer. The purposes of the proposed method are automating manual keyword annotation and achieving high speed and high accuracy keyword annotation. Experimental results showed that the proposed method is more accurate than TFIDF and Naive Bayes in P@5 and P@10. Moreover, these results also showed that the proposed method could annotate about 19 times faster than Naive Bayes.