In 2022, the global online travel market amounted to as much as 474.8 billion dollars, estimated to exceed one trillion by 2030. Economic growth, cultural exchange, technological advancements, digital transformation, and emerging markets are some factors that contributed to the growth of the travel industry.
As the market experiences a significant surge, the need for improved customer service has intensified, and this demand requires a closer look at customer reviews.
A prominent online travel agency joined forces with iMerit to handle reviews classification, a critical task for online travel companies. This process requires categorizing and scrutinizing customer reviews and feedback, covering every aspect of the travel experience, ranging from the quality of travel services to the provided accommodations.
About the Client
The client is among the leading online travel accommodation platforms and promotes responsible travel practices. The company offers a comprehensive platform for individuals and businesses to plan and book travel arrangements.
The Challenges
Review classification is critical for businesses in the travel sector to understand customer sentiments, identify primary areas of improvement, and make data-driven decisions. Like any other big online travel platform, our client faced a significant challenge in managing data from three distinct sources of reviews. Each data source has a unique set of labels, creating a complex web of overlapping information. For example, the customer reviews dataset had 200+ unique labels, partner hub comments had 10+ labels, and travel community posts had 40+ labels.
Handling a multitude of labels across diverse datasets posed a significant challenge for the client. With strategic objectives focused on swift and informed decision-making, they faced the daunting task of managing this extensive data within a tight two-month deadline. This time-sensitive scenario demanded a meticulous approach to ensure data was efficiently organized and ready for insightful analysis.
The Solution
Recognizing the intricate nature of the challenge, iMerit devised a comprehensive solution that leveraged the combined expertise of Subject Matter Experts (SMEs) and Natural Language Processing (NLP) consultants. The SMEs played a crucial role in discerning domain-specific data intricacies, while the NLP consultants assisted in developing a nuanced understanding of the textual data and the relationship between different labels.
Through a collaborative effort between NLP experts and SMEs, our team successfully grouped labels into 38 distinct categories strategically designed for efficiency and client needs.
The Result
- Faster Scaling: By categorizing topics into 38 distinct labels, we successfully accelerated the data processing, aligning with the client’s timeline for scalable and sustainable data operations.
- 98.5% Accuracy: The labeling process achieved an impressive average team accuracy of 98%, ensuring datasets were contextually relevant. This high level of accuracy directly translated into more precise analyses, providing actionable insights that were reliable and impactful.
Conclusion
Data annotation for review classification involves accurately labeling or tagging reviews based on their content or sentiment, and annotation quality annotation is crucial for ML models to learn effectively.
While many companies rely on automated tools or crowdsourcing platforms for annotation, human supervision or verification is essential to maintain annotation quality. Our collaborative approach categorized diverse data sources into 38 distinct labels, allowing our client to process data within their timeline while ensuring highly relevant and strategically aligned datasets.