Scaling Spreading Activation for Information Retrieval

摘要

Further analysis of the scaling properties of the production IRIA system are underway, using additional queries and topics and additional data sets with larger harvest values. While this work is ongoing, we can point to ways to address the limitations this research has uncovered in applying spreading activation to information retrieval. This could include revised parameterizations which lower the propagation threshold or increase the initial activation of certain terms, or revised network structures which take high-fanout words related to user interests and break them apart into lower-fanout concepts using structures such as redundant discrimination nets.