| บทคัดย่อ |
งานวิจัยนี้พัฒนานวัตกรรมการท่องเที่ยวเชิงอัจฉริยะ โดยใช้วิธีการผสานการเรียนรู้ของเครื่อง (Machine Learning) ร่วมกับโครงสร้างฐานข้อมูลแบบ Star Schema เพื่อแนะนำการท่องเที่ยวเชิงวัฒนธรรมและการท่องเที่ยวเชิงอาหารที่นักท่องเที่ยวชื่นชอบในจังหวัดเพชรบูรณ์ มีวัตถุประสงค์
1) เพื่อศึกษารูปแบบความต้องการของนักท่องเที่ยวต่อแพลตฟอร์มดิจิทัลอันชาญฉลาดเพื่อยกระดับเศรษฐกิจการท่องเที่ยวเชิงอาหารจังหวัดเพชรบูรณ์ และ2) เพื่อพัฒนาแพลตฟอร์มดิจิทัลอันชาญฉลาดเพื่อยกระดับเศรษฐกิจการท่องเที่ยวเชิงอาหารจังหวัดเพชรบูรณ์ โครงสร้าง Star Schema ช่วยจัดการข้อมูลอย่างมีประสิทธิภาพ โดยตาราง Fact_FoodTourism เชื่อมโยงนักท่องเที่ยว อาหาร และแหล่งท่องเที่ยวเข้ากับตารางมิติ Dim_Tourist, Dim_Food และ Dim_TourismDestination ซึ่งสามารถดึงข้อมูลได้ เช่น คะแนนอาหารที่ชื่นชอบ การให้คะแนนแหล่งท่องเที่ยว ข้อมูลประชากรนักท่องเที่ยว และส่วนผสมอาหาร นอกจากนี้ยังมีการคำนวณระยะทางด้วยสูตร Haversine และกรองนักท่องเที่ยวที่อยู่ในรัศมี 5 กม. เพื่อสร้างบริบทสำหรับคำแนะนำแบบเฉพาะบุคคล สนับสนุนการทำนายความสนใจและความชอบของนักท่องเที่ยวผ่านการทำ Clustering, Classification และ Recommendation Models ข้อมูลถูกรวบรวมจากนักท่องเที่ยว 384 คน ผ่านแบบสอบถามตามกรอบแนวคิด 5A (Attraction, Accessibility, Accommodation, Activities, Amenities) และสัมภาษณ์ผู้เชี่ยวชาญด้านอาหาร/ผู้ประกอบการท้องถิ่น 15 คน เพื่อเก็บข้อมูลเชิงคุณภาพ คำตอบแบบไม่เป็นโครงสร้างถูกประมวลผลด้วย NLP เพื่อนำมาสร้างชุดข้อมูลที่เป็นโครงสร้าง จากนั้นใช้ SQL Queries บน Star Schema เพื่อสกัดคุณลักษณะเข้าสู่การเรียนรู้ของเครื่อง ผลการทำ K-Means Clustering พบว่า จำนวนกลุ่มที่เหมาะสมที่สุดคือ 9 กลุ่ม (Best K = 9) โดยมีการแยกกลุ่มปานกลาง (Silhouette Score = 0.491, Calinski-Harabasz = 323.48, Davies-Bouldin = 0.716) นักท่องเที่ยวสูงอายุ งบประมาณสูง (Clusters 0 และ 4) มักเดินทางไกลและสนใจหลากหลาย ขณะที่นักท่องเที่ยวอายุน้อย งบประมาณจำกัด (Clusters 2 และ 5) มักเลือกเดินทางใกล้ ๆ ปัจจัยเพศก็มีผลต่อความชอบ: กลุ่มที่มีผู้หญิงมากให้ความสำคัญกับความพึงพอใจด้านอาหารและแหล่งท่องเที่ยว ส่วนกลุ่มผู้ชายให้ความสำคัญกับการเดินทางไกลหรือใช้จ่ายมากกว่า สำหรับโมเดลการพยากรณ์ ประเภท Tree-based (Random Forest, XGBoost) และ Neural Networks (MLP) ทำได้แม่นยำถึง 100% ในการทำนายหมวดหมู่อาหารที่ชื่นชอบ, K-Nearest Neighbors ทำได้ 98.7%, ขณะที่ Logistic Regression ได้ 64.9% ด้านการแนะนำ (Recommendation) พบว่า Collaborative Filtering ให้ผลดีกว่าอัลกอริทึมอื่น ๆ
ผลการวิจัยพบว่า การจัดการข้อมูลเชิงโครงสร้างร่วมกับการใช้ Machine Learning และ Recommendation Algorithm ขั้นสูง สามารถสร้างคำแนะนำด้านการท่องเที่ยวได้อย่างแม่นยำและเฉพาะบุคคล ช่วยเพิ่มประสิทธิภาพในการออกแบบแพ็กเกจท่องเที่ยวเฉพาะกลุ่ม การวางแผนการเดินทาง การตลาดเชิงกลยุทธ์ และเพิ่มความพึงพอใจของนักท่องเที่ยวโดยรวม
This study develops an intelligent tourism innovation using an integrated machine learning approach based on the star schema for recommending tourists’ favorite cultural and culinary tourism in Phetchabun Province. The purposes are 1) To study the patterns of tourists’ demands for an intelligent digital platform to enhance the culinary tourism economy in Phetchabun Province. And 2) To develop an intelligent digital platform to enhance the culinary tourism economy in Phetchabun Province. The star schema efficiently organizes data, with Fact_FoodTourism linking tourists, foods, and destinations to Dim_Tourist, Dim_Food, and Dim_TourismDestination. Attributes such as favorite food scores, destination ratings, tourist demographics, and food ingredients can be retrieved. Distance calculations using the Haversine formula, filtered for tourists within 5 km, provide context for personalized recommendations, supporting the prediction of tourists’ preferences and interests through clustering, classification, and recommendation models. Data were collected from 384 tourists via a questionnaire based on the 5A framework (Attraction, Accessibility, Accommodation, Activities, Amenities), and 15 food experts/local entrepreneurs were interviewed for qualitative insights. Unstructured responses were processed using NLP to create a structured dataset. SQL queries on the star schema extracted features for machine learning. K-Means clustering identified nine tourist groups (Best K = 9) with moderately well-separated clusters (Silhouette Score = 0.491, Calinski-Harabasz = 323.48, Davies-Bouldin = 0.716). Older, higher-budget tourists (Clusters 0 and 4) traveled longer distances with diverse interests, while younger, budget-conscious tourists (Clusters 2 and 5) favored shorter trips. Gender influenced preferences: female-dominated clusters prioritized satisfaction with food and destinations, while male-dominated clusters preferred longer trips or higher spending. Tree-based models (Random Forest, XGBoost) and neural networks (MLP) achieved 100% accuracy in predicting preferred food categories; K-Nearest Neighbors reached 98.7%, while Logistic Regression scored 64.9%. For recommendations, collaborative filtering outperformed other algorithms. These results demonstrate that combining structured data management with advanced machine learning and recommendation algorithms can deliver accurate, personalized tourism recommendations, enhancing tailored travel packages, optimized itineraries, strategic marketing, and overall tourist satisfaction.
|
| คำสำคัญ |
การเรียนรู้ของเครื่อง,การแนะนำ,การท่องเที่ยวเชิงวัฒนธรรมและการท่องเที่ยวเชิงอาหาร,แบบจำลองดาว,Machine Learning,Recommendation,Cultural and Culinary Tourism,Star Schema
|