—This paper proposes a method for extracting various opinions about evaluation points from web reviews. In
the proposed method, the evaluation points which frequency
is different for each evaluation value such as stars. Opinions
in reviews with extremely high or low evaluation values are
effective to evaluate the review targets in most cases. However,
the reviewers themselves are diverse. The evaluation points that
many reviewers highly evaluated sometimes can be the cause of
the low evaluation in a specific context for the same target. It is
thus reasonable to say that comparing those conflicted opinions
should help us to understand the features of the services and
products. To provide such findings to the readers, focusing on
not only the evaluation points which has high frequency but
also various opinions in different context should be required.
In this paper, we conducted the extraction experiment and had
the discussion about the extracted opinions.