| 作成者 |
|
|
|
|
|
| 本文言語 |
|
| 出版者 |
|
| 発行日 |
|
| 収録物名 |
|
| 巻 |
|
| 開始ページ |
|
| 終了ページ |
|
| 会議情報 |
|
| 出版タイプ |
|
| アクセス権 |
|
| 関連DOI |
|
| 関連DOI |
|
|
|
| 概要 |
When conducting a search for research papers, the search should return comprehensive results related to the user’s query. In general, a user inputs a Boolean query that reflects the information need, ...and the search engine ranks the research papers based on the query. However, it is difficult to anticipate all possible terms that authors of relevant papers might have used. Moreover, general query-based ranking methods emphasize how to rank the relevant documents at the top of the results, but require some means of guaranteeing the comprehensiveness of the results. Therefore, two ranking methods that consider the comprehensiveness of relevant papers are proposed. The first uses a topic-based Boolean query search. This search converts every word in the abstract set and query into a topic via topic analysis by Latent Dirichlet Allocation (LDA) and conducts a search at the topic level. The topic assigned to synonyms of a search term is expected to be the same as that assigned to the search term. Each paper is ranked based on the number of times it is matched with each topic-based Boolean query search executed for various LDA parameter settings. The second is a hybrid method that emphasizes better results from our topic-based ranking result and a general query-based ranking result. This method is based on the observation that the paper sets retrieved by our method and by a general ranking method will be different. Through experiments using the NTCIR-1 and -2 datasets, the effectiveness of our topic-based and hybrid methods are demonstrated.続きを見る
|
Mendeley出力
