Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/15970
Title: | 搜尋結果多樣性問題中,基於二項分布子議題滿足度之研究 A Study on Binomial-distribution-based Subtopic Satisfaction Model for Search Result Diversification Problem |
Authors: | Ching-Han Tzou 鄒京翰 |
Advisor: | 鄭卜壬 |
Keyword: | 搜尋結果多樣性,子議題新穎度, Search Result Diversity,Subtopic Novelty, |
Publication Year : | 2012 |
Degree: | 碩士 |
Abstract: | 為了提高同時滿足不同使用者的機率,搜尋結果多樣化試圖讓搜尋結果適度涵蓋搜尋關鍵字的各種意圖與面相,亦即子議題。隨著網路資訊量逐年成長及冗餘資訊的增加,搜尋結果多樣化的地位愈趨重要。近年來愈來愈多研究者投入此一研究,相關的會議與競賽也十分熱絡。本研究詳細分析了數個現有的搜尋結果多樣化方案及相關的效能評估方法,認為子議題新穎度為目前研究的重點。針對搜尋結果多樣化,作者提出了一個新的子議題新穎度模型,藉由一個二項分配機率模型模擬使用者掃描搜尋結果、尋找有用資訊之行為。透過預測使用者對一特定子議題的需求,在閱讀相關文件後所得到的滿意度,可以進一步推測子議題的新穎度變化。
本研究使用文字檢索會議中搜尋結果多樣性競賽之資料集,並與現有方法比較。實驗結果顯示本研究所提出之新穎度二項機率模型能夠進一步提昇目前一名列前茅之現有方法 xQuAD 的效能。此外,實驗結果亦指出本方法具有能夠適應各種不同性質搜尋詞的能力,可以針對不同搜尋詞提出最適宜的新穎度估計方法。 Search result diversification aims to satisfy different types of user at the same time by providing a proper mixture of interpretations and aspects of a single query string, i.e. query subtopics. As the Web grows steadily, it has been more and more important today; competitions have been hold and researches have been carried out in recent years. In this thesis, with several state-of-the-art methods and evaluations being studied thoroughly and observations made, a new subtopic novelty model that improves diversification effectiveness by modeling user satisfaction with a binomial random process is proposed. It models subtopic novelty by estimating the satisfaction a user attains when he or she skims over a result document. Results of experiments on the TREC Diversity Task competition dataset show that the proposed novelty model further improves one of the most dominating method today, namely xQuAD. Experiments show that it has the capability to be fine-tuned for queries with different attributes as well. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/15970 |
Fulltext Rights: | 未授權 |
Appears in Collections: | 資訊工程學系 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-101-1.pdf Restricted Access | 2.35 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.