由於資訊的爆炸，已造成資訊的供大於求，使用者須經過不斷的瀏覽及搜尋才能找到所喜歡的商品，因此許多學者開始鑽研資訊過濾的機制，此機制除了解決資訊爆炸的問題外，還能向使用者推薦符合其需求的資訊，幫助使用者能夠過濾並選擇適合自己的商品，而推薦系統就是屬於資訊過濾的一種應用，能夠依據使用者的喜好、需求及興趣，將資訊或商品推薦給使用者，減少使用者在搜尋過程中付出額外時間成本。目前常見的推薦系統形式有內容式過濾、協同式過濾，以及結合上述兩種的混合式，目前混合式推薦已被廣泛應用於電子商務業界。 為求推薦系統所推薦之項目能更符合人們意願，即推薦之項目更為人們接受，因此本研究目的提出一個結合網路輿情的電子商務推薦系統（以下簡稱結合網路輿情型），於背景收集使用者紀錄，透過自動化擷取、意見單元定義擷取與極性分析等步驟收集討論區網友的評論，對商品進行網路輿情分析，給予商品分數並進行商品推薦，最後以結合網路輿情型與未結合網路輿情的傳統型電子商務推薦系統（以下簡稱傳統型）比較其使用者使用意願與推薦準確率來評估本研究之成果。 本研究利用問卷調查、收集使用者紀錄進行推薦系統評估方法與結合網路輿情型的推薦排名影響使用者意願得知，近八成民眾滿意結合網路輿情型，且其所推薦結果之F-Measure為70.48%較傳統型高出近15.75%，可以得知結合網路輿情型之推薦結果較符合使用者心中意願，且其推薦準確率達到九成，且結合網路輿情型會影響使用者對商品的喜好排名。 Due to the information explosion which causes the overload of information, the users have to continuously browse and search to get the information they preferred. Therefore, many scholars start to study information filtering to reduce the problem of information explosion and help user to select the items based on user’s preferences. Recommendation system is one of the use of the information filtering which can provide information or item based on user’s preferences, requirements and interests more efficiently and precisely. Currently, the recommendation methods can be categorized into the following three types: content-based, collaborative and hybrid recommendation methods. The hybrid recommendation method has been most widely used in the field of e-commerce. In order to meet the needs of the user when select products from the recommendation system, this study therefore presented the e-commerce recommendation system with sentiment analysis (ECRS-SA). By using automatic data extraction, opinion extraction, and polarity analysis to collected user’s usage records and comments from the social networks. Then rate the products according to the results of sentiment analysis and give recommendation to users. Finally, we compared ECRS-SA and ECRS-SA without Sentiment Analysis (ECRS) to evaluate user’s need and recommendation accuracy. In this paper, we used the survey of satisfaction and the recommendation system evaluated method to analyze user’s usage records. The experiment results show that eighty percent of people satisfy with the ECRS-SA and the F-Measure of the recommendation result is 70.48% higher than the ECRS by 15.75%. The ECRS-SA results is much match to user’s need and the accuracy rate is over ninety percent. And the user’s preference will be impacted by the ECRS-SA.
