The Markov Retrieval Model Based on Transfer Learning

摘要

Traditional information retrieval model is trained on static dataset, which might be violated when a task from new domain comes. In this paper, we propose a Markov retrieval model based on transfer learning. Our solution is to first construct a retrieval model based on Markov Network on old data, and then employ new data to update old Markov Network. Moreover, the distances between different distributions are measured using the KL divergence. The one-step and multi-step transferring experiments on TREC datasets prove the model can transfer well among several datasets. Multi-step transferring experiment indicates that the learning ability of model has nothing to do with the learning sequence.