1 / 25

Navi 下一步工作的设想

Navi 下一步工作的设想. 郑 亮 6.6. LOD Cloud. Knowledge Graph. Motivation. When browsing an entity or a set of entities in the Semantic Web, it is important to improve the efficiency of human navigation and help people find the information they need as fast as possible.

ryanadan
Télécharger la présentation

Navi 下一步工作的设想

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Navi 下一步工作的设想 郑 亮 6.6

  2. LOD Cloud

  3. Knowledge Graph

  4. Motivation • When browsing an entity or a set of entities in the Semantic Web, it is important to improve the efficiency of human navigation and help people find the information they need as fast as possible. • Our study aims to not only recommending the related entities, but also understanding how people reach these entities by navigating through the Semantic Web.

  5. Relate d Work • Navigating in Information Networks • User click-trail analysis • Network structure analysis

  6. Recommended Papers • West R, Paranjape A, Leskovec J. Mining missing hyperlinks from human navigation traces: A case study of wikipedia. www2015: 1242-1252. • West R, Leskovec J. Human wayfinding in information networks. www2012: 619-628. • Blanco R, Cambazoglu B B, Mika P, et al. Entity Recommendations in Web Search. ISWC2013. • Antikacioglu A, Ravi R, Sridhar S. Recommendation Subgraphs for Web Discovery. www2015.

  7. 1. Human Wayfinding in Information Networks [West 2012] Stanford University • Task:How do humans navigate information networks? • How to do? • Understand how humans navigate Wikipedia Get an idea of how people connect concepts. • Study more than 30,000 goal-directed human search paths and identify strategies people use when navigating information spaces. • Apply the lessons learned, in order to design a learning algorithm for predicting an information seeker’s target, given only a prefix of a few clicks.

  8. Example path

  9. Data collection via a game:Wikispeedia.net http://cs.mcgill.ca/~rwest/wikispeedia/ More than 30,000 instances (the data came from around 9,400 distinct IP addresses).

  10. How good are humans at finding such short chains?

  11. Elements of Human Wayfinding • Anatomy of typical paths • Making progress is easiest far from and close to the target. • Hubs are crucial in the opening. • Conceptual distance to the target decreases steadily • Big leaps first, followed by smaller steps. • Clicks are most predictable far from and close to the target • Two main strategic elements • Degree-based: Navigate to hub • Similarity-based: Get ever closer to target in terms of semantic distance

  12. Target Prediction • Our next goal is to apply the lessons learned, in order to design a learning algorithm for predicting an information seeker’s target, given only a prefix of a few clicks. • Our method explicitly takes the characteristic features of human search into account and is trained on real human trajectories. • We cast our task as a ranking problem. Given the observed path prefix q, rank all articles t according to how plausible they are as targets of the current search.

  13. Learning-to-rank model

  14. 2. Mining Missing Hyperlinks from Human Navigation Traces[West 2015]

  15. Task • Navigation logs for mining missing links. • If we often observe users going through page s and ending up in page t, although s does not directly link to t, then it might be a good idea to introduce a ‘shortcut’ link from s to t.

  16. Ranking by relatedness • It seems reasonable to rank source candidates s by their relatedness to t, since clearly a link is more relevant between articles with topical connections • Ranking by path frequency

  17. 3. Entity Recommendations in Web Search [Blanco, 2013] • Task • Given the large number of related entities in the knowledge base, we need to select the most relevant ones to show based on the current query (entity) of the user. • Approach • Entity Recommendation task  Ranking task • For every triple in the knowledge base, Spark extracts over 100 features (co-occurrence, popularity, and graph-theoretic features(PageRank),…).

  18. 3. Entity Recommendations in Web Search [Blanco, 2013]

  19. 4. Recommendation Subgraphs for Web Discovery [Antikacioglu 2015] Carnegie Mellon University … … L: the set of discovered items R: the set of undiscovered

  20. Our Approach T : the related entities path • User click-trail  Meta-Paths • Learning to rank … s : current entity l: the length of path

  21. Thanks! • Q&A

More Related