search

Trogluddite / loombreaker

Tools for building Topic-Specific Web Indexes (CS-480 Capstone)
MIT License
0 stars 0 forks source link

explore mechanics for improving the relevance /recency of the Bayeseian network #44

Open Trogluddite opened 9 months ago

Trogluddite commented 9 months ago

Documents on the web change, links die, etc.

How do we update a Bayesian network in-place to reflect these changes?

One idea:

  1. each document has a 'weight' datastructure, composed of: a. list of tuples it contributes to the network b. count of each tuple
  2. upon removal, we decrement the tuples by the appropriate amount, essentially 'removing' it from the bayesein network.
  3. possible challenge: if we're using evolutionary mechanisms to retrain the Bayesian network, these weights will become dynamic.