Determining Directions of Service Quality Management using Online Review Mining with Interpretable Machine Learning

• This paper has been published in the International Journal of Hospitality Management (IJHM), April 2024.

Explanation

• Identifying service features and estimating their importance are crucial for service management.
• We propose a systematic framework for estimating the importance of service features using online review mining with interpretable machine learning.
• LDA and VADER are utilized to extract service features and calculate the sentiment (satisfaction) of each service feature from online reviews, respectively.
• For estimation of feature importance, machine learning models and SAGE are used to learn and interpret the nonlinear relationship between service features satisfaction and overall customer satisfaction.

Implementation

• Our Environment
  • OS: Ubuntu 18.04 (we checked this code file works on Windows OS)
  • python == 3.7
  • conda >= 4.10

Run this code to create your virtual environment. (you can skip it, but we recommend creating a new virtual environment for this implementation)

conda create -n 'env_name' python=3.7 

Activate your environment

conda activate 'env_name'
# (or) source activate 'env_name'

Run this code to clone the files into your local environment.

git clone https://github.com/SQwithIML/OnReviewServImprovement.git

Move to the OnReviewServImprovement folder

cd OnReviewServImprovement/

Run the below code to install the required package for this implementation.

#(if you need) pip install --upgrade pip
pip install -r requirements.txt

Code structure

root
├─Params.yaml
├─identifyServiceFeatures.py
├─estimateImportance.py
├─src
│   ├─ __init__.py
│   ├─ main_components.py
│   ├─ preprocessors.py
│   └─ utils.py
├─Datasets
│   └─ Custom
│      └─***.** (filename.extension)
└─Results
    ├─ [1]LDA - Result_Folder_for_each_combination
    ├─ [2]ML - Hyper_parameter_tuning_Reports
    ├─ [3]IPA - Plot_for_Importance_Performance_Analysis
    └─ etc.

Note

Before you run the below code, please check the Params.yaml file. In Params.yaml, you can set the parameters for our framework and custom data path. Since a grid search will be performed to find optimal LDA and ML models, you have to set the candidates of each hyperparameter of both models. Also, you can change the name of .yaml file. If you change it, please enter the file name at --ymal="ymal_name" in the below codes.

Due to the copyright, we cannot provide the review data that we employed for our paper. Instead, we provide open-source review data for this implementation. To implement this framework for your review datasets (star rating required), just modify ModifiedCustomizedPreprocessor in identifyServiceFeatures.py to fit your datasets, and change USE_DATATYPE to "Custom" and CUSTDATA_PATH in Params.yaml . For this, please refer to the example code in ModifiedCustomizedPreprocessor.

[1] Review data preprocessing and service feature identification

• Execution list
  • Text preprocessing
  • LDA Topic modeling

Running the code below generates preprocessed datasets in /Results folder and a summary.txt and report.txt with a words-topic list for each combination of candidates in each folder in /Results/[1]LDA/.

python identifyServiceFeatures.py --yaml="Params"

Referring to the summary.txt and words-topic list in report.txt, choose the best results of LDA topic modeling and name each topic. After that, Enter the folder name at LDA_RESULT_IDX (for stage 2) and the name of each topic in TopicList in the yamlfile. Note that the number of topics in the 'TopicList' must match the number of topics of the best LDA model you choose.

[2~3] Global importance estimation using the optimal prediction model

• Execution list
  • Preparing the datasets for training (creating a review-feature matrix and splitting datasets)
  • K cross-validation for hyper-parameter tuning and finding optimal prediction model
  • Estimation of the global importance of service features
  • Importance Performance Analysis (plot)

Running the code below generates report.txt that records the performance of each model in /[2]ML folder and saves the plot for Importance Performance Analysis (IPA) in /[3]IPA folder.

python estimateImportance.py --yaml="Params"

Citation Information

APA

Shin, J., Joung, J., & Lim, C. (2024). Determining directions of service quality management using online review mining with interpretable machine learning. International Journal of Hospitality Management, 118, 103684.

Bibtex

@article{shin2024determining,
  title={Determining directions of service quality management using online review mining with interpretable machine learning},
  author={Shin, Jongkyung and Joung, Junegak and Lim, Chiehyeon},
  journal={International Journal of Hospitality Management},
  volume={118},
  pages={103684},
  year={2024},
  publisher={Elsevier}
}