In electronic commerce, most of customers refer product reviews
that are written by previous purchasers to decide which product they
buy or whether they buy or not a certain product. As many reviews
are provided on internet shopping malls, we can f...
In electronic commerce, most of customers refer product reviews
that are written by previous purchasers to decide which product they
buy or whether they buy or not a certain product. As many reviews
are provided on internet shopping malls, we can find out various and
objective evaluation on each product, but it is not easy to analyze bulk
of reviews. Most of internet shopping malls present purchaser’s
preference to each product with asterisk scores, but we cannot decide
whether we buy it or not because that score includes too various
information of a product such as quality, price and delivery. A lot of
researches on product reviews in English have been studied, but they
need a large amount of knowledge resources like Word so it is not
relevant to apply their methods to Korean. Recently some studies on
Korean product reviews are attempted but those require knowledge
that is constructed by hand. In this thesis, I propose an automatic
product review summarization system. The system discriminates
whether a product review is positive or negative per product
evaluation items, which are features of a product such as color, price,
size and delivery. Product reviews are abstracted with polarity scores
based on evaluation items, so users can easily catch evaluations of
previous purchasers without reading bulk of reviews. The proposed
system consists of 4 steps: 1) collecting and refining reviews , 2)
extracting evaluation items using term statistics in reviews and web
information retrieval, 3) extracting evaluation words and their polarity
per each evaluation item using natural language processing technique,
4) presenting polarities of each evaluation item graphically. All steps
only use automatically extracted knowledge from reviews and web. In
experiment using reviews from online shopping malls and my system
shows 90.3% in extracting evaluation items when comparing with
correct answers made by hand and get 7.6 out of 10 in polarity
resolution.
，韩语论文网站，韩语论文范文 |
