以食物名詞辨識為基礎的菜單建立與食物推薦系統

Abstract

Web 2.0是一種新的網際網路方式，將使用者視為網路應用的中心。在Web 2.0的網站中，使用者除了單方面被動的瀏覽及搜尋資訊外，也可以自由地發表文章與表達意見，透過這些參與及互動，使網路資源變得更豐盛。現今有越來越多的網路平台提供使用者分享生活經驗，例如對於旅遊、餐廳、音樂、電影等等之評論。而隨著Web 2.0的盛行，人們也習慣於透過手持式裝置及無線網路，隨時隨地搜尋其他人所發表的文章以得到一些有用的資訊並作為參考。然而，使用者往往必須花費非常多的時間及精神來閱讀並消化這些大量的文章。 由於許多人都有在到某家餐廳用餐前，花很多時間到網路上閱讀他人食記以得知哪些食物較為好吃的經驗，我們提出了一個可以自動分析、匯整大量食記而建立出食物推薦清單及餐廳相關資訊的系統，來節省使用者的時間，並作為點餐時的參考。在此論文中，部落格探勘的方法，首先與名詞辨識做結合，以抽取食記中出現的食物名稱，進一步產生食物清單，接下來再用情緒分析及社群排序計算每道食物的排序分數。經過實驗的結果顯示，系統回傳的食物推薦清單對使用者有實質上的幫助，而其中食物名稱辨識的準確率也令人滿意。

Web 2.0 is an important concept of World Wide Web and users are considered to be the core in Web 2.0 applications. In addition to browsing and retrieving information passively and unilaterally, users can post articles and share opinions freely on a Web 2.0 site. There are lots of platforms enabling users to publish and share their life experiences, such as travel, restaurant, and music. As the popularity of Web 2.0, people are getting used to search for user-generated data from the internet anytime and everywhere through handheld devices and universal high speed 3G mobile networks to get useful information and refer to others’ experiences. However, it is time-consuming to read and digest a large amount of data by ourselves. To alleviate the problem of spending a lot of time to read lots of restaurant reviews for figuring out which food is delicious, we propose a system which can return food recommendation list and restaurant related information automatically by blog mining technique. The blog mining method is first combined with food name extraction algorithm we proposed to generate food name list, and then sentiment analysis and social ranking are applied to calculate the ranking score of each food. To evaluate the performance of the system, three experiments are conducted. From the results of the experiments, we find that the food recommendation list is helpful for users and the accuracy of food name extraction is satisfactory.