In this paper, we designed a novel method to extract comparative
relations from customer opinion data, to build comparative relation
maps for aiding enterprise managers in identifying the potential
operation risks and supporting strategy decisions. The two-level CRF
model with unfixed interdependencies can better extract the
comparative relations, by utilizing the complicated dependencies
between relations, entities and words, and the unfixed interdependencies
among relations. The empirical evaluation demonstrated the
effectiveness of this model. The comparative relation map is
potentially a very effective tool to support enterprise risk management
and decision making. The contributions of this paper include the
following: 1) To the best of our knowledge, this is the first work on
using comparison opinion as information sources in CI for enterprise
risk management; 2) the proposed graphical model can achieve better
performance for relation extraction by modeling the unfixed interdependencies
among relations, which is not covered by the existing
methods; and 3) the empirical evaluation shows that the performance
of the comparative relation extraction is quite promising, and it
implies the feasibility of mining the comparison opinions for CI.