텍스트 마이닝을 이용한 온라인 리뷰의 레스토랑 유형별 속성에 관한 연구

다국어 초록 (Multilingual Abstract)

The hospitality industry, including the restaurant industry, is enhancing its
competitiveness by using the core technology of the 4th industrial revolution.
As consumers' service experiences can be written in real time, related review
data is continuously accumulating. In particular, in the restaurant industry, due
to the intangibility, which is a characteristic of services, consumers refer to
online reviews are essential. In other words, consumers can minimize risks and
uncertainties in the consumption process through online reviews, and
information obtained through reviews is used to select and evaluate restaurants.
At this time, the restaurant attributes serve as a criterion for comparing
restaurants and making purchase decisions in the purchase decision process,
and at the same time affect customer satisfaction. In particular, there is a
difference in restaurant attributes because consumers' needs vary depending
on the type of restaurant. Therefore, this study aims to classify restaurants into
quick service, casual dining, and fine dining types to identify differences in
customer restaurant attributes and analyze what attributes are important for
each type of restaurant. TripAdvisor reviews were collected using Python for
the United States, which has the largest restaurant industry. Based on the
collected review data, word frequency analysis was performed by deriving the
simple frequency and TF-IDF values of words for each restaurant type using a
text mining technique. In addition, through LDA analysis, a topic modeling
method, the contents of reviews by restaurant type can be classified into which
topics, and the differences in topics by restaurant type were compared. First,
the difference in words mentioned by restaurant type was confirmed through
word frequency analysis. In the case of quick service restaurants, keywords
related to food, price, service speed, and location frequently appeared. In the
case of casual dining restaurants, the words about the employees providing
service and the atmosphere appeared at the top, and in the case of the fine
dining restaurant, the words expressing the service, atmosphere, and purpose
of visit frequently appeared. Second, through topic modeling LDA analysis, all
three restaurant types were classified into four topics. Overall, food and service
attributes are essential, indicating that they are the most important attributes to
consider in the process of selecting and evaluating restaurants regardless of
type. In addition, it shows that the more complex the type of restaurant that
requires services, the more complex the restaurant attributes are considered.
Through this study, the difference in restaurant attributes was confirmed by
understanding the needs of customers according to the restaurant type, and it
provided useful implications for establishing management and marketing
strategies according to the restaurant type.

국문 초록 (Abstract)

외식산업을 비롯한 Hospitality산업은 4차 산업혁명의 핵심 기술을 이용하여 경쟁력을 제고하고 있다. 소비자의 서비스 경험을 실시간으로 작성할 수 있게 되면서 이와 관련된 리뷰 데이터가 지속적으로 축적되고 있다. 특히, 레스토랑 산업은 무형성이라는 서비스의 특성으로 인해 소비자는 간접적인 경험을 제공하는 온라인 리뷰를 필수적으로 참고하게 된다. 즉, 소비자는 온라인 리뷰를 통해 소비 과정에서의 위험과 불확실성을 최소화할 수 있게 되었고, 리뷰를 통해 얻어지는 정보를 레스토랑을 선택하고 평가하는 데 활용하고 있다. 이때, 레스토랑 속성은 구매의사결정과정에서 레스토랑을 비교하고 구매를 결정하는 데 기준이 되고, 동시에 고객의 만족도에 영향을 미친다. 특히, 레스토랑의 유형에 따라 소비자의 니즈가 달라지기 때문에 레스토랑 속성에도 차이가 있다. 따라서 본 연구는 레스토랑을 퀵서비스, 캐주얼다이닝, 파인다이닝 유형으로 구분하여 고객의 레스토랑 속성의 차이를 확인하고, 각 유형의 레스토랑마다 중요하게 여기는 속성이 무엇인지 분석하고자 한다. 이를 위해 레스토랑 산업의 규모가 가장 큰 미국을 대상으로 파이썬을 활용하여 트립어드바이저의 리뷰를 수집하였다. 수집된 리뷰 데이터를 기반으로 텍스트 마이닝 기법을 이용하여 레스토랑 유형별 단어의 단순빈도와 TF-IDF 값을 도출하여 단어 빈도분석을 시행하였다. 또한, 토픽모델링 방법인 LDA분석을 통해 레스토랑 유형별 리뷰의 내용이 어떠한 토픽으로 구분될 수 있는지 확인하고, 레스토랑 유형별 토픽의 차이를 비교하였다. 첫째, 단어 빈도분석을 통해 레스토랑 유형별 언급되는 단어의 차이를 확인하였다. 퀵서비스 레스토랑의 경우 주로 음식과 가격, 서비스 속도, 위치에 관련한 키워드가 자주 등장했다. 캐주얼다이닝 레스토랑의 경우 서비스를 제공하는 종업원과 분위기에 대한 단어가 상위에 나타났고, 파인다이닝 레스토랑의 경우 주로 서비스와 분위기, 방문 목적을 나타내는 단어가 자주 나타났다. 둘째, 토픽모델링 LDA 분석을 통해 세 가지 레스토랑 유형 모두 4개의 토픽으로 분류하였다. 전반적으로 음식과 서비스 속성은 필수적으로 포함되어 유형에 상관없이 레스토랑을 선택하고 평가하는 과정에서 가장 중요하게 고려하는 속성임을 알 수 있다. 또한, 서비스가 더욱 요구되는 레스토랑의 유형일수록 레스토랑 속성이 더욱 복합적으로 고려된다는 점을 보여준다. 본 연구를 통해 레스토랑 유형에 따른 고객의 니즈를 이해함으로써 레스토랑 속성의 차이를 확인하였고, 레스토랑 유형에 따른 경영 및 마케팅 전략을 수립하는 데 유용한 시사점을 제공하였다.

목차 (Table of Contents)

• Ⅰ. 서론 ································· 1
• 제1절 연구의 배경 ································· 1
• 제2절 연구의 목적 ································· 5
• 제3절 연구의 방법 ································· 6
• 제4절 연구의 구성 ································· 7
• Ⅱ. 이론적 배경 ································· 9
• 제1절 레스토랑 속성 ································· 9
• 1. 레스토랑 속성 ································· 9
• 제2절 레스토랑 온라인 리뷰 ································· 15
• 1. 온라인 리뷰 ································· 15
• 2. 레스토랑 온라인 리뷰 ································· 20
• 제3절 텍스트 마이닝 ································· 23
• 1. 텍스트 마이닝의 개념 ································· 23
• 2. 텍스트 마이닝 과정 ································· 25
• 3. 텍스트 마이닝 선행연구 ································· 27
• Ⅲ. 연구 방법 ································· 30
• 제1절 연구대상 및 데이터 수집 과정 ································· 30
• 1. 연구대상 ································· 30
• 2. 데이터 수집 ································· 33
• 3. 데이터 전처리 ································· 34
• 3.1 토큰화(tokenization) ································· 34
• 3.2 불용어(stopwords) 제거 ································· 34
• 3.3 정규화(normalization) ································· 35
• 제2절 연구모형 ································· 36
• 1. TF-IDF 분석 ································· 36
• 2. LDA(Latent Dirichlet Allocation) ································· 38
• Ⅳ. 연구 결과 ································· 41
• 제1절 데이터 수집 결과 ································· 41
• 제2절 레스토랑 유형별 단어 빈도분석 ································· 43
• 1. 단어 빈도(TF) 분석 결과 ································· 43
• 2. TF-IDF 분석 결과 ································· 49
• 제3절 레스토랑 유형별 LDA 분석 ································· 56
• 1. 토픽 개수 선정 ································· 56
• 2. 토픽 추출 및 결과 ································· 56
• Ⅴ. 결론 ································· 65
• 제1절 결론 및 시사점 ································· 65
• 제2절 한계점 및 제언 ································· 68