DESIGNING A TEXT MINING METHOD TO DISCOVER HOTEL BRAND PERSONALITY TRAITS BASED ON CONSUMER ONLINE REVIEWS

Outline

1. Background of the Problem

2. Problem Identification and Formulation

3. Assumptions and Problem Limitations

4. Research purposes

5. Research methodology

6. Data processing

7. Conclusions and recommendations

​

​

Background of The Problem

Brand personality is a framework that is useful for companies in increasing the value of products, services, and missions to society.

This is because through reviews from consumers, hotels can evaluate hotel brands and assess customer satisfaction.

According to Amatulli (2017), consumers are currently concerned about social and environmental issues.

According to Gimenez (2012), companies or hotels must focus on positive social, environmental and financial aspects during the process.

Definition of Brand Personality (BP)

The Importance of Online Consumer Reviews

Changes in Consumer Behavior

Sustainability in Hotel

To make a good hotel brand personality, hotels need data from user reviews using the text mining method.

Text Mining Method

​

​

Problem Identification and Formulation

Frequency of Traits Each Dimension

Radar Chart

Disadvantages of Previous Research

Formulation of the Problem

1. What is the method for finding traits based on the word embedding and word2vec methods from the data review you have?

2. What is the method for determining the importance of the sustainability dimension to other dimensions that already exist in Aaker (1997)?

3. How is the performance obtained from the proposed text mining method?

4. What are the proposed improvements to the dimensions of the brand personality of the hotel used as a case study, namely Hotel Lotte New York Palace based on the sentiment analysis method?

Limitation of Problems and Assumptions

1. The research was conducted using data from the Tripadvisor website for the Lotte New York Palace hotel from 2013 to 2022 and the RIU Plaza Hotel from 2016 to 2022.

2. This data was obtained from consumer reviews on the Tripadvisor website for the Lotte New York Palace and RIU Plaza hotels.

3. The assumption used in this study is that the data is taken once from hotel reviews on Tripadvisor, not from repetition.

Research purposes

1. Design a method to find traits based on the text mining method and use the reviews you have.

​

2. Designing a method to determine the importance of the sustainability dimension to other dimensions that already exist in Aaker.

3. Calculating the performance obtained from the current text mining method.

4. Obtain proposals for repairing the Lotte New York Palace Hotel based on the results of the sentiment analysis method.

Problem Identification and Formulation

Limitation of Problems and Assumptions

Determination of Research Objectives

Literature review

Data collection

Trait Search (Word Embedding)

Determination of Dimensional Importance (Dependency Parsing and Tree)

Performance Measurement Methods

​

Proposal to Improve Hotel Lotte New York Palace (Sentiment Analysis)

Analysis

Conclusions and recommendations

Research Methodology

​

​

Data processing

​

​

Data Collection

Preparation of Lotte New York Palace Hotel Dataset

Data Preprocessing

Data Training

Determine the Trait Words to be Given

Use Word2vec & Wordnet Models

Show Results

Trait Search

​

​

Preprocessing Data

​

​

​

​

Data Training

Vektor [0,1, 0,3, 0,4,..]

“second visit hotel”

​

​

Most Similar Words

(Word2vec & Wordnet)

​

​

Most Similar words “Honest” Trait

W2Vec : involved, alone, after, stated, embarrassing, lost, together, angry, through, dan continued.

Wordnet : honest, honorable, dependable, reliable, true, good, dan fair.

P (involved & honest ) = 0.98

….

….

involved, honest, dependable, reliable, alone, after, stated, embarrassing, lost, together, angry, through, continued, true, good, and fair.

Comparison of word2vec and wordnet synonyms

Selected Synonyms

​

​

Honest Trait Frequency

Sincerity Dimension

Honest Trait

involved, honest, dependable, reliable, alone, after, stated, embarrassing, lost, together, angry, through, dan continued, true, good, dan fair.

Involved = 8

alone = 9

……

“Honest “ Trait =1422

​

​

Sincerity Dimension Frequency

Sincerity Dimension

honest = 1422 sincere = 168

real = 379

original = 227

cheerful = 181

Friendly = 2559

Sincerity Dimension = 4936

​

​

Frequency Each Dimension

​

​

Dependency Parsing & Dependency Tree

Token Head

( Hotel, Room, Staff)

Hotel

Most similar words “Hotel”, “Staff”, “Room”

lotte, property, place, palace, decorations, exterior, interior, gym, architecture, bedroom, bed, windows, toiletries, bathroom, mattress, furniture, rooftop, guestroom, doormen, bellmen, concierge, members, employees, bartenders, bellman, doorman, captain, maids, housekeeping.

​

​

Result

(Parent – Child)

​

​

​

​

Result

(Parent – Child)

Dependency Sibling

​

​

Performance

(Parent – Child)

• The results of the dependency parsing process produce 1968 review sentences that have context with hotels.
• From 1968, there are 5 sentences of review that have no context with the hotel.

“especially the bar was a real highlight (golden room) and we had the opportunity to try some of the signature cocktails.”

​

“The outside courtyard facing the backside of the catherdral was lovely and a pleasant place to relax and have coffee.”

​

“when you walk into the courtyard from the street, you feel like you've been transported to a different place & time.”

“we had a great time staying here - the hotel has a lovely courtyard that is a really peaceful and comfortable place to relax and unwind after a busy day.”

​

“we had a great time staying here - the hotel has a lovely courtyard that is a really peaceful and comfortable place to relax and unwind after a busy day.”

Performance

(Siblings)

The results of the dependency parsing sibling process produce 42 review sentences that have context with the hotel.

​

​

Uncountable Example Sentences by Dependency Parsing (Lotte NY Hotel)

​

​

Difference in Parent Child Dependency Parsing Frequency

RIU Hotel

• The results of the parent child dependency parsing process produce 1335 review sentences that have context with the hotel.

• The results of the dependency parsing sibling process produce 3 review sentences that have no context with the hotel.

1. “the smaller breakfast room was more relaxing than the larger "canteen”.

2. “it's a big busy train station ...no personal service.”

3. “chocolates in the room on my birthday lovely and a cake at breakfast with the whole room singing happy birthday something i won't forget.”

​

​

Sentiment Analysis

detect the use of words that indicate negative, neutral or positive sentiments.

Sentiment scores are on a scale of -1 to +1, where -1 indicates negative sentiment, +1 indicates positive sentiment, and 0 indicates neutral sentiment.

There are 4 scores: compound, negative, neutral, positive

​

​

Sentiment Analysis

​

​

Sentiment Analysis

​

​

Sentiment Analysis

​

​

Sentiment Analysis

These results can provide insight for the Lotte New York Palace Hotel that the majority of reviews on the dimension of sincerity at the Lotte New York Palace Hotel have a positive value. So the Lotte New York Hotel needs to maintain the hotel's sincerity qualities such as friendly and courteous service to all guests, providing solutions to complaints or questions raised by guests.

​

​

Sentiment Analysis

In the excitement dimension, most of the reviews are positive hotel reviews, so there are several things that need to be maintained in terms of the excitement dimension, such as places and hotel facilities that make customers comfortable.

​

​

Sentiment Analysis

On the competence dimension, negative reviews tend to have more frequency than the previous dimension, but the majority are positive. There are a number of things that the hotel can pay attention to based on these negative reviews, such as price adjustments in the future based on current hotel facilities

​

​

The positive reviews on the sophistication dimension are very high, this indicates that the hotel has good sophistication quality. The hotel's calm and comfortable atmosphere is a positive review of the most sophistication dimensions that this hotel has

Sentiment Analysis

​

​

Sentiment Analysis

Reviews of the ruggedness dimension are quite good, where the majority of visitors rate it as positive. However, there are suggestions to improve the consistency of hotel service quality. Because some customers receive poor treatment from hotel staff

​

​

Sentiment Analysis

On the sustainability dimension, the majority of reviews received are positive. The peaceful atmosphere of the hotel, along with the beautiful views around the hotel, make customers satisfied with this hotel. So it needs to be maintained at this point.

​

​

Text Mining Method Design

Hotel Review Data

Trait Search (Word embedding)

Frequency Across Dimensions

Analysis using Dependency Parsing & Dependency Tree

Frequency Across Dimensions

Sentiment Analysis

Proposed Improvements

​

​

Conclusion

1. The word embedding method is used to find traits using word2vec. This method looks for the most similar words from each trait. Then, to minimize the poor level of accuracy, a similarity process is carried out using the word2vec method with synonyms from wordnet. If the probability of similarity is more than 0.5 then the word is taken, and vice versa.

2. The dependency parsing & dependency tree method is used to determine the importance of the hotel's brand personality dimension which aims to analyze sentences based on context. In this research, we tested the dependency parsing parent child and dependency parsing sibling. Dependency parent child produces more information than parsing sibling dependencies. By using this method, it is possible to assess the sentence or word resulting from which traits have a relationship or context with the hotel.

​

​

Conclusion

​

3. Based on the performance of the dependency parsing method currently used, the level of performance is very good. At the Lotte New York Palace hotel, using dependency parsing parent child, it produces 1968 review sentences, and only 5 review sentences that are not quite right. Then, using dependency parsing sibling, produces 42 review sentences with very good accuracy, then research at the RIU Plaza hotel, using parent child dependency parsing, produces 1335 review sentences and only 3 review sentences that are not quite right.

​

​

Conclusion

​

4. Proposed improvements to the hotel's brand personality based on sentiment analysis for the sincerity dimension are that Hotel Lotte New York needs to maintain the quality of hotel sincerity such as friendly and polite service to all guests, even in the excitement dimension the hotel needs to maintain excitement quality such as a comfortable place and hotel facilities , then on the competence dimension, hotels need to adjust prices in the future based on the hotel facilities they have, on the sophistication dimension, hotels need to maintain a comfortable hotel atmosphere, on the ruggedness dimension, hotels need to improve the consistency of hotel service quality, and on the sustainability dimension, hotels need to preserve the view which is well located around the hotel.

​

​

recommendation

1. It is necessary to carry out further information extraction processes using other NLP methods such as developing NLP for chatbots.

2. Analyzing texts in different languages, not only English.

3. Implement word embedding in a different domain: Apart from word2vec, there are many other word embedding techniques that can be used to represent text.

4. Conduct deeper research to increase research recall.

​

​

Thank You

​

​

Data Preprocessing (1)

​

​

Preprocessing (2)

​

​

Most Similarity

​

​

Sinonym Frequency

​

​

Library

​

​

Sentiment Analysis

​

​

Sentiment Analysis