Informax principle-based query expansion using Hopfield neural networks

摘要

Asymmetric HNN designed as an associative memory for query expansion has been researched in some papers. However, there is no criterion in this method to measure its validity and to tell good results from bad ones objectively. What's more, convergence characteristic of HNNs may not be guaranteed if the symmetry is broken. Aiming at avoiding these two points, maximum mutual information (informax) principle-based query expansion using symmetric HNNs is proposed from the perspective of combinatorial optimisation.