Real-Time Semantic Matching and Recommendation Redirected from Real-Time Semantic Matching and Recommendation

Real Time Semantic Matching and Recommendation (1D) 

Convener: Marcus Trevisani from  BINTRO/ BINTRO works on match-making in an abstract sense between people/jobs/information 

Notes-taker(s):  

A. Tags for the session - technology discussed/ideas considered:

B. Discussion notes, key understandings, outstanding questions, observations, and, if appropriate to this discussion: action items, next steps: 

 Agents/assistants in semantic web 

Semantic web and real-time notification 

Challenges of real-time web 

Retrieval/analysis/association/use of semantic technologies.  

Scaling - NxN problem - need to divide and conquer.  

Usage of ontology’s 

geonames (location), synonyms, domain-specific (e.g.) structure of industries 

Building of ontology’s: Some ontology’s are established and years in the making. Others are built on the fly based on natural-language and other semantic techniques. Open ontology’s would be nice. 

Real-time  

It was pointed out that "real-time" means different things in different contexts. It does not always imply that information is delivered in seconds. It could emphasize the asynchronous nature of the delivery.  

Active vs. passive model of searching 

How is RT search different from traditional search? Fewer results with higher relevance, continuous results.  Semantic analysis for a search query might require more time than conventional searches based on pre-computed indices and trade-off delivery time for quality of the result.  BINTRO, forms to enter data, matching happens on events (update and creation of forms) based on semantic modeling 

Freshness/Relevancy 

Michael Jackson death was discussed as an example: Evolving story, RTW can aggregate and prioritize to give the right information at any given time. Content feeds generate events, popularity/frequency are tracked on ongoing basis. Key is indexing.  

How is trust/authority addressed?  

Options include using known (static) sources and using social/user reputation mechanisms.  User benefit in the double-blind style match-making to prevent one side from getting spammed. For example, in Craigslist the advertiser can get irrelevant contacts whereas the RTW can help by interjecting an "agent".