在臺灣餐飲業以無人監督機器學習進行顧客分類

Abstract

食品消費的無形和體驗性使顧客尋求在線評論來評估餐廳質量。然而，大多數市場細分過程涉及使用成本和時間密集型消費者調查來確定細分基礎，很少有研究探索使用社交媒體和其他平台上現有的在線評論。該研究提供了一種新穎的方法，用於根據真正代表客戶聲音的實際評論來細分市場，而不是試圖根據先前研究論文中預先設想的客戶滿意度維度來細分客戶，並幫助克服設計和設計的挑戰和努力託管傳統流程，同時有效減少當前客戶研究方法的偏差、不確定性和成本。 本研究利用稱為潛在狄利克雷分配 (LDA) 過程的無監督機器學習算法從台灣 Yelp 平台上的 6541 條在線客戶評論中識別九個客戶體驗維度。本次台灣餐飲業研究發現的顧客體驗維度是“等待時間”、“氛圍”、“響應和禮貌”、“物有所值”、“傳統食物”、“菜單選擇”、“餐廳佈局” , “烹飪風格”, “食物味道和風味”。接下來，我們使用這九個客戶體驗維度，使用 K-means 聚類將餐飲市場客戶分為四個客戶群，然後使用評論情緒得分、星級評分、門店類型和美食等數據對這些客戶群進行分析。發現的客戶群被命名為“美食愛好者”、“傳統”、“不參與”和“有意識的消費者”。該研究在基於已發現客戶體驗維度的已發現客戶細分分析的基礎上，進一步做出了重要的理論、方法和實踐貢獻。此外，所提出的細分方法可以幫助解決根據客戶偏好細分市場的需求，並且可以自動化，與調查不同，定期進行時，可以捕捉新興的消費者偏好並儘早為營銷人員揭示有價值的見解，證明元素在設計適當的分割目標定位。

Food consumption's intangible and experiential nature makes customers seek online reviews to evaluate the restaurant quality. However, most market segmentation processes involve identifying the segmentation bases using cost and time-intensive consumer surveys, with few studies exploring the use of online reviews existing on social media and other platforms. The study provides a novel methodology for segmenting a market based on actual reviews that genuinely represent the voice of the customer rather than trying to segment customers based on preconceived customer satisfaction dimensions done in previous research papers and helping to overcome the challenges and effort of designing and hosting conventional processes while effectively reducing current customer research methods' bias, uncertainty, and cost. This study utilizes an unsupervised machine learning algorithm called as Latent Dirichlet allocation ( LDA) process to identify nine customer experience dimensions from 6541 online customer reviews on the Yelp platform for Taiwan. The found customer experience dimensions in this study of the restaurant industry in Taiwan were “Waiting time”, “Atmosphere”, “Responsiveness and courtesy”, “Value for money”, “Traditional food”, “Menu selection”, “Restaurant layout”, “Cooking style”, “Food taste and flavor”. Next, we used these nine customer experience dimensions to segment the food-outlets market customers into four customer segments using K-means clustering, who later were profiled using data of review sentiment scores, star ratings, type of outlets, and cuisines. The discovered customer segments were titled as “Food lovers”, “Traditional”, “Uninvolved”, and “Conscious consumer”. The study further makes significant theoretical, methodological, and practical contributions based on profiling of discovered customer segments based on found customer experience dimensions. Furthermore, the proposed segmentation method can help address the need to segment a market based on customer preferences and can be automated which when conducted periodically, unlike surveys, can capture emerging consumer preferences and reveal valuable insights for marketers early on, proving elemental to be in designing appropriate segmentation targeting positioning.