PaperlayerView exact revision on arXiv

Source-linked AI summary

A systematic literature review on machine learning applications for consumer sentiment analysis using online reviews

Praphula Kumar Jain, Rajendra Pamula

arXiv:2008.10282v1cs.HC

TL;DR

The paper addresses the challenge of extracting meaningful information from large, subjective, heterogeneous online-review data for consumer sentiment analysis. It systematically reviews machine-learning applications in hospitality and tourism, finding that machine learning is implemented for consumer sentiment analysis and that detection is a major application, while reinforcement-learning applicability remains a future possibility.

  • Problem

    Large, subjective, and heterogeneous social-web data make it challenging to obtain meaningful information manually for consumer sentiment analysis.

  • Method

    The paper systematically reviews 68 articles on machine-learning applications in consumer sentiment analysis using online reviews in hospitality and tourism.

  • Results

    Detection is identified as a major machine-learning application, and machine learning is implemented for consumer sentiment analysis in hospitality and tourism using online reviews.

  • Takeaways & Limitations

    The review organizes existing machine-learning applications and evidence to address research queries and inform future research in this area.

  • Takeaways & Limitations

    The review identifies the applicability of reinforcement learning in consumer sentiment analysis as a possibility for future work.

Abstract

from arXiv · show

Consumer sentiment analysis is a recent fad for social media related applications such as healthcare, crime, finance, travel, and academics. Disentangling consumer perception to gain insight into the desired objective and reviews is significant. With the advancement of technology, a massive amount of social web-data increasing in terms of volume, subjectivity, and heterogeneity, becomes challenging to process it manually. Machine learning techniques have been utilized to handle this difficulty in real-life applications. This paper presents the study to find out the usefulness, scope, and applicability of this alliance of Machine Learning techniques for consumer sentiment analysis on online reviews in the domain of hospitality and tourism. We have shown a systematic literature review to compare, analyze, explore, and understand the attempts and direction in a proper way to find research gaps to illustrating the future scope of this pairing. This work is contributing to the extant literature in two ways; firstly, the primary objective is to read and analyze the use of machine learning techniques for consumer sentiment analysis on online reviews in the domain of hospitality and tourism. Secondly, in this work, we presented a systematic approach to identify, collect observational evidence, results from the analysis, and assimilate observations of all related high-quality research to address particular research queries referring to the described research area.

1. Introduction

The introduction frames machine learning as a way to process growing, heterogeneous online-review data for consumer sentiment analysis in hospitality and tourism. It identifies a literature gap and positions the review as a systematic synthesis of relevant applications and evidence.

  • Machine learning is used to process online reviews because their increasing volume and unstructured form make manual extraction of meaningful information difficult.
  • Online reviews provide consumer experiences that organizations can use to improve service quality and policies, while prospective consumers can consult them before purchasing.
  • Consumer sentiment analysis extracts sentiments, opinions, and demands from domain-specific reviews and identifies their polarity.
  • Prior literature examined machine-learning applications for sentiment analysis across several domains, but no survey had reviewed online-review applications in hospitality and tourism.
  • The review includes 68 articles addressing machine-learning applications for sentiment classification, predictive recommendation decisions, and fake-review detection.
  • The study aims to help researchers understand the current role of machine learning in hospitality-and-tourism sentiment analysis and support organizational growth.

2. CSA: Process, technniques, Challanges

This section describes consumer sentiment analysis as extracting polarity and experience-related information from online reviews, then outlines machine-learning categories, algorithms, and major analytical challenges. It emphasizes subjectivity, context dependence, short texts, fake reviews, and computational cost as obstacles to reliable processing.

  • Consumer sentiment analysis: Consumer sentiment analysis determines whether reviews are negative, positive, or neutral and can identify attitudes toward services using words, tone, and context.
  • Consumer sentiment analysis: Organizations can use sentiment findings to understand consumer satisfaction, improve services, and identify issues in newly released products.
  • CSA process: The described CSA workflow includes data-processing steps, feature-selection techniques, and a process diagram for extracting information from reviews.
  • Challenges: Subjectivity is difficult to identify because the same review may appear neutral to some consumers but not others, with interpretations varying across consumers and languages.
  • Challenges: Short texts often lack contextual information, making subjectivity classification difficult and increasing the likelihood of false reviews or misleading classifications.
  • Challenges: Context dependency means that a word such as “long” can be positive for a battery but negative for a queue, requiring broader sentence structure for interpretation.
  • Challenges: Processing large vocabularies makes training computationally costly and time-consuming, while fake and unopinionated reviews can confuse sentiment classification.
  • Machine-learning techniques: Machine-learning algorithms are grouped into supervised, unsupervised, and reinforcement learning, with examples including decision trees, clustering, neural networks, and Q-learning.

3. Systematic Literature Review (SLR) Methodology

The paper uses a three-stage systematic literature review to survey machine-learning applications for consumer sentiment analysis in hospitality and tourism. It searches and filters English online-review studies using defined keywords and quality criteria, then validates the final selection through author re-evaluation.

  • Review design: The review follows a systematic literature review methodology designed to survey published research accurately and maintain an unbiased summary of recent work and future directions.The process is framed as a scientific methodology for evaluating research related to the review topic.
  • Review design: The study applies a three-stage SLR process covering review planning, conducting the review, and reviewing its results.
  • Search strategy: The search uses broad consumer-sentiment terms combined pairwise with machine-learning terms and hospitality-and-tourism applications.Keywords include Naive Bayes, Support Vector Machine, Neural Network, regression, consumer sentiment, consumer opinion, and recommendation prediction.
  • Selection criteria: The review excludes studies without justifiable contributions or benchmarks, non-text review data, reviews without findings, and non-English consumer sentiment analysis.Image, video, and audio data were not considered.
  • Quality validation: Backward and forward searching, author re-evaluation, and agreement-based coding produced a final selection of 68 articles for the review.Articles judged unsuitable after re-evaluation were excluded, and the process emphasized article relevance rather than journal quality ratings alone.

4. Review findings and discussions

The review organizes machine-learning applications in hospitality and tourism into sentiment analysis, predictive recommendation, and fake-review detection. It finds ML useful for extracting sentiment and opinions from online reviews, supporting consumer and organizational decisions.

  • Review findings and discussions: The SLR divides consumer sentiment-analysis applications into sentiment analysis, predictive recommendation, and fake-review detection.Each sub-area serves a particular application in the consumer-sentiment-analysis process.
  • Review findings and discussions: ML is implemented for consumer sentiment analysis in hospitality and tourism management using online reviews.
  • Predictive recommendation decision: ML techniques support predictive recommendation decisions based on previous consumer experiences with offered services.
  • Fake-review detection: Fake-review detection identifies reviews that may misguide consumers before purchase decisions.
  • Sentiment analysis: Sentiment analysis classifies consumer feelings or emotions as positive, negative, or neutral using text analysis, ML, and NLP.Models can identify consumer satisfaction toward products, services, or offerings in online reviews.
  • Review findings and discussions: ML can capture textual features without requiring high-level feature extraction.

7 KNN 5 [18], [20], [30], [31], [58]

The reviewed literature applies multiple machine-learning techniques to consumer reviews, especially in hotel, airline, and tourism settings. It identifies gaps in review coverage and recommends more advanced predictive approaches for future work.

  • Sentiment analysis: Most datasets in the reviewed sentiment-analysis studies are hotel, airline, and tourism reviews.Airport and museum reviews remain future research opportunities.
  • Review findings and discussions: Consumer reviews can inform service improvement, consumer policies, and forthcoming purchase decisions.
  • Predictive recommendation decision: NPS is calculated as the percentage difference between promoters and detractors and is used as a consumer-satisfaction measure.
  • Predictive recommendation decision: The research gap concerns automatically extracting review information to predict recommendation drivers, understand service advice, and form it into NPS.
  • Predictive recommendation decision: The reviewed studies use regression and other basic machine-learning techniques for predictive recommendation decisions.Future work may apply ensembling and optimization procedures.
  • Fake-review detection: A fake review may be positive or negative and can promote a seller’s products or damage competitors’ sales.

1. To improve the products ranking

The review presents machine learning as useful for extracting consumer sentiment from online reviews and supporting classification, prediction, recommendation, and fake-review detection in hospitality and tourism. It proposes an ML-CSA framework linking review collection, preprocessing and visualization, machine-learning techniques, and consumer sentiment analysis.

  • Applications: The SLR finds machine learning applicable to sentiment analysis, predictive recommendation, and fake-review detection using hospitality and tourism reviews.The reviewed data from various online platforms is used for classification and predictive recommendation decisions.
  • Findings: Machine-learning techniques improve consumer-sentiment-analysis accuracy and address challenges in sentiment analysis, fake-review detection, and predictive recommendation decisions.The review also reports that machine learning helps identify factors responsible for consumer sentiment.
  • ML-CSA framework: The proposed ML-CSA framework has four phases: online-review collection, preprocessing and visualization, machine-learning techniques, and consumer sentiment analysis.The framework begins with data selection aligned to the research goal and includes sources such as hotel, airline, restaurant, airport, and tourist reviews.
  • Implications: The framework positions decision making as the final stage of the consumer sentiment analysis process and is intended as guidance for future research.The authors describe the framework as a bridge between consumer sentiment analysis and machine-learning techniques.

1 RMSE

This section lists performance measures used in the reviewed studies, including RMSE, precision, recall, F-measure, and accuracy.

  • Classification measures: Precision, recall, F-measure, and accuracy are among the performance measures applied across the selected articles.The listed formulas define these measures using true-positive, false-positive, and false-negative quantities.

10 AUC 3 [5], [29], [30]

The review identifies implications for researchers and service providers while outlining language, data, tooling, and methodological gaps in consumer sentiment analysis. It reports that core service aspects influence sentiment more than augmented aspects and that sentiment can inform predictive recommendations.

  • Implications for researchers: The review aims to guide researchers in evaluating consumer sentiment analysis with machine-learning techniques.The proposed framework is presented as a guideline for this research area.
  • Research gaps: The SLR is based primarily on English-language online reviews, leaving reviews in languages such as Hindi, Parsi, Urdu, and Bengali for further study.The review explicitly states that no attention was paid to written reviews in another language.
  • Research gaps: Only a few studies used optimization algorithms, ensemble learning, deep learning, or neuro-fuzzy models in consumer sentiment analysis.The authors identify these techniques as possible directions for future work.
  • Research gaps: Most reviewed studies analyzed qualitative online reviews, while relatively few combined qualitative and quantitative review content.The review also identifies limited attention to regional and cultural preferences and museum-review datasets.
  • Implications for service providers: Consumer choices depend on region, culture, cost, and facilities, requiring service policies to attend to these factors.The review presents context as relevant when designing advertisements and service policies.
  • Implications for service providers: Sentimental words related to joy are reported to have more impact than words related to surprise or trust.The finding is presented as a consideration for advertisement content.

6. Summery

This systematic literature review synthesizes machine-learning applications for consumer sentiment analysis of online reviews in hospitality and tourism and proposes an ML-CSA framework. It reports benefits across sentiment classification, recommendation prediction, and fake-review detection, while identifying scope and validation limitations.

  • Scope and method: The SLR examines recent studies covering data selection, data sources, preprocessing, feature selection, machine-learning techniques, objectives, and future scope.The review is presented as a systematic approach to understanding the research area.
  • Scope and method: The review covered 68 research papers on machine learning for consumer sentiment analysis.The reported review period spans January 2017 to July 2020.
  • Findings: Supervised and unsupervised machine-learning techniques are used for sentiment analysis, predictive recommendation, and fake-review detection.These applications are identified specifically in hospitality and tourism online-review research.
  • Contribution: The proposed ML-CSA framework provides guidelines ranging from topic selection and data preparation to publication-related information.The authors present it as a novel contribution intended to help new and experienced researchers.
  • Implications: The review concludes that implementing machine learning in consumer sentiment analysis is beneficial for identifying consumer-satisfaction factors and supporting business growth.Service providers may use online-review findings to draw factors related to consumer satisfaction.
  • Limitations: The SLR may have missed important articles, excludes some conference papers, and is limited to selected journals and databases.The authors recommend including top conferences and additional databases in future studies.
  • Limitations: The proposed ML-CSA framework has not been empirically tested and requires validation in future work.The framework is based on the review findings rather than an empirical evaluation.

Appendix A. Journals description from which articles selected for SLR

Appendix A lists journals from which the review articles were selected, including their quartiles, indexing services, and citation statistics where available. It also records publication-related declarations and funding information.

  • Journal coverage: The appendix identifies journals across Q1 and Q2 quartiles, indexed in SSCI or SCIE, including Journal of Business Research, Information Processing and Management, and Information & Management.For these three journals, minimum, average, and maximum citation values are reported as 273/311/348, 178/252/325, and 211/314/416, respectively.
  • Journal coverage: Additional selected journals include Multimedia Tools and Applications, Sustainability, Knowledge-Based Systems, Journal of Computational Science, and Annals of Tourism Research.Their reported average values are 236, 50, 270, 166, and 135, respectively.
  • Table structure: The appendix table is organized by serial number, journal name, quartile, indexing service, and minimum, average, and maximum values.The table continues from a previous page, and several entries contain unavailable values.
  • Publication declarations: The article declares no conflict of interest, reports no human or animal studies, and states that it received no specific grant funding.The ethical-approval statement is split across two passages.
Loading 2008.10282v1…