| 摘要: | 情感分析屬於自然語言處理領域中的分支，主要目的為判斷評論者對於產品或服務的回饋是屬於正向還是負向情感。自從社群網路的崛起，越來越多人願意在平台上分享產品使用心得或是服務經驗，成為決策者新的參考依據。餐廳業者也設計線上問卷來收集顧客的用餐滿意度，藉此改善各缺點以提升顧客回流率。隨著評論的增加，很難以人工的方法瀏覽評論，需透過電腦取代人力來取得寶貴的評論資訊。目前情感分析的應用大多屬於文檔級，依據全部的評論預測出一個情感極性，忽略了評論可能包含對多個觀點的意見及情感。
為了讓使用者查看評論中針對各個觀點的情感極性，透過歷史顧客評論幫助使用者快速了解餐廳各方面的優缺點，本研究將情感觀點分析用於餐廳評論。首先使用詞嵌入將語句和觀點的字詞轉化為詞向量作為電腦計算的輸入來源，第二將觀點詞彙分為食物、價格、服務、氣氛及軼事五大類，第三使用觀點嵌入之長短期記憶結合注意力模型依照給定的觀點標籤判斷評論語句為正向、負向或中立情感。結果顯示該模型的預測準確率高達84%。藉由情感觀點分析的結果可建餐廳評論分類系統，將應用資料集區分成食物、價格、服務、氣氛等用餐體驗四大構面及軼事，使用者透過系統分別查看關於餐廳不同觀點的顧客評論，依據情感極性的分類結果得知餐廳在不同觀點上的顧客滿意度。情感觀點分析用於餐廳評論將與日俱增的評論中分門別類，對於餐廳業者及顧客都能快速找到有用的資訊。
;Sentiment analysis belongs to a branch in the field of natural language processing. The main purpose is to determine whether the feedback of the reviewer to the product or service is positive or negative emotion. Since the rise of social networking, more and more people are willing to share product experience or service experience on the platform, becoming a new reference for decision makers. Restaurants have also designed online questionnaires to collect customers′ dining satisfaction, thereby improving various shortcomings and increasing customer return rates. With the increase of reviews, it is difficult to browse the entire review by manual methods. Therefore, it is necessary to use computers to replace manpower to obtain valuable comment information. At present, most of the applications of sentiment analysis are document level which the sentiment polarity is predicted based on all the comments, ignoring that the comments may contain opinions and sentiments on multiple aspects.
In order to allow users to view the sentiment polarity of each aspect of view in the review and understand the advantages and disadvantages of the four aspects of the restaurant′s dining experience, this study uses aspect based sentiment analysis for restaurant reviews. First, use word embedding to convert the word of sentences and aspects into word vectors as the input source of computer calculation. Second, divide the aspect word into five categories: food, price, service, atmosphere and anecdotes. Third, use Attention-based LSTM with Aspect Embedding model judges the comment sentence as positive, negative or neutral emotion according to the given five categories label. The results show that the prediction accuracy of the model is as high as 84%. A restaurant review classification system can be built based on the results of aspect based sentiment analysis. The application data set can be divided into food, price, service, atmosphere and other four aspects and anecdotes. Reviews, based on the classification results of emotional polarity, found the restaurant’s customer satisfaction from different viewpoints. Emotional opinion analysis is used in restaurant reviews to classify the ever-increasing reviews, so that restaurant owners and customers can quickly find useful information. |
