Personalized concept-based search on the Linked Open Data

详细信息    查看全文

• 作者：Melike Sah^a ; ^{melike.sah@neu.edu.tr" class="auth_mail" title="E-mail the corresponding author} ; Vincent Wade^b ; ^{Vincent.Wade@scss.tcd.ie" class="auth_mail" title="E-mail the corresponding author}
• 关键词：Categorization ; Concept-based search ; Fuzzy retrieval model ; Personalized search/exploration ; Linked Open Data ; UMBEL
• 刊名：Web Semantics: Science, Services and Agents on the World Wide Web
• 出版年：2016
• 出版时间：January 2016
• 年：2016
• 卷：36
• 期：Complete
• 页码：32-57
• 全文大小：5547 K

文摘

In this paper, we present a novel personalized concept-based search mechanism for the Web of Data based on results categorization. The innovation of the paper comes from combining novel categorization and personalization techniques, and using categorization for providing personalization. In our approach, search results (Linked Open Data resources) are dynamically categorized into Upper Mapping and Binding Exchange Layer (UMBEL) concepts using a novel fuzzy retrieval model. Then, results with the same concepts are grouped together to form categories, which we call conceptlenses. Such categorization enables concept-based browsing of the retrieved results aligned to users’ intent or interests. When the user selects a concept lens for exploration, results are immediately personalized. In particular, all concept lenses are personally re-organized according to their similarity to the selected lens. Within the selected concept lens; more relevant results are included using results re-ranking and query expansion, as well as relevant concept lenses are suggested to support results exploration. This allows dynamic adaptation of results to the user’s local choices. We also support interactive personalization; when the user clicks on a result, within the interacted lens, relevant lenses and results are included using results re-ranking and query expansion.

Extensive evaluations were performed to assess our approach: (i) Performance of our fuzzy-based categorization approach was evaluated on a particular benchmark (∼10,000 mappings). The evaluations showed that we can achieve highly acceptable categorization accuracy and perform better than the vector space model. (ii) Personalized search efficacy was assessed using a user study with 32 participants in a tourist domain. The results revealed that our approach performed significantly better than a non-adaptive baseline search. (iii) Dynamic personalization performance was evaluated, which illustrated that our personalization approach is scalable. (iv) Finally, we compared our system with the existing LOD search engines, which showed that our approach is unique.