

社群媒體中顧客知識之挖掘:意見探勘技術開發
22
consumer reviews, the class association rule mining algorithm is applied to learn a set of
opinion sentence identification rules for the target product feature. The identification rules
can be employed to distinguish between opinionated sentences and non-opinionated
sentences of the target product feature.
3. Findings
To evaluate the effectiveness of the proposed R-OSI technique, a data set comprising of
4,500 consumer review sentences regarding digital cameras was collected from Amazon and
Google Shopping. We manually labeled each review sentence as containing zero or more
product features. Across the whole data set, we identified eight product features: battery,
flash, image quality, lens, memory, price, screen, and video; for which 149, 83, 273, 241, 95,
156, 102, and 67 sentences were respectively identified as opinion sentences. A supervised
learning approach proposed by Yang et al. (2010) was adopted as our performance
benchmark. A class association rule mining algorithm was also adopted as its underlying
induction learning method. According to our empirical evaluation results, the R-OSI
technique, with one keyword for each product feature, achieved comparable performance
with the fully supervised benchmark approach, in which the manual preparation of training
data is necessary. Furthermore, when we increased the number of keywords for each product
feature from one to three, the performance of the R-OSI technique improved appreciably.
The macro- and micro-F-measure values increased by 5.4% and 4.6%, respectively. We also
investigated the sensitivity of the R-OSI technique to the size of the set of unannotated
consumer reviews. The proposed technique demonstrated stable performance for different
sizes of unannotated consumer review corpora. Overall, these findings suggest that the
proposed R-OSI technique achieves promising performance in opinion sentence
identification, even when a supervised learning approach is adopted as the performance
benchmark.
4. Research Limitations/Implications
In addition to theoretical contributions to the literature on opinion mining, our study
offers several practical implications. First, social media can be a crucial resource for
companies to explore and develop business intelligence. However, understanding consumer-
oriented business insights embedded in the considerable amount of online consumer reviews
can be highly challenging. Effective opinion mining techniques or tools can increase the