Debbie Rank Typed Graph Walks Presentation Joshua Shinavier 1

From Semantic Portal Wiki

• Question is for the Presentation: Debbie Rank Typed Graph Walks Presentation
• Question is asked by: Joshua Shinavier
• The Question is:[[Question_asked:=

The authors imply that the superiority of the reranking method to the weight-tuning method is due to its "global" nature, as opposed to the "local" nature of the latter method. They go on to explain that the weight-tuning method does not model multi-step dependencies. However, I believe that this is a limitation of the chosen technique and not of lazy walk algorithms in general. A slight variation of the presented technique would add some notion of a path type (e.g. the "sent-to", "sent-from-inverse" edge sequence mentioned by the authors) to the configuration, and direct activation spreading according to those types. See this paper for an example of such an approach.]]