3,727
reviews analyzed
Degree thesis, Universidad de Manizales, 2026
Sentiment Analysis in Hotel Reviews of Buenaventura through Text Mining
The dashboards on this site come from approved academic research, published in the institutional repository of Universidad de Manizales. Here you'll find the full document (in Spanish), how to cite it, and what it found.
1,710
of them with text, the basis of the sentiment analysis
2020 - 2025
period analyzed
90.2%
agreement with human review
The gap it filled
Buenaventura is positioning itself as a tourist destination: in January and February 2025 alone it received more than 31,000 visitors, drawn by humpback whale watching, the beaches of Juanchaco and Ladrilleros, and the San Buenaventura carnivals. At the same time, it carries a public image marked by violence and reinforced by media stigma, which weighs on the confidence of anyone deciding whether to travel there.
Thousands of those visitors had already written down their experience, but no one had organized it. Before this work, no study had rigorously systematized the opinions published about the city's hotels, so the evidence of what they got right or wrong remained scattered across individual comments. This research built that corpus, then turned it into comparable indicators.
Results by hotel
Digital reputation indicators reported in the document for the 2020 to 2025 period. Each hotel has its own page with the thematic detail and its year-by-year evolution.
| Hotel | Reviews | Rating | Positive | Negative | Neutral | Answered |
|---|---|---|---|---|---|---|
| Hotel Cordillera | 1,193 | 3.94 | 47.48% | 34.53% | 17.99% | 0.92% |
| Hotel Cosmos Pacífico | 788 | 4.46 | 63.37% | 27.72% | 8.91% | 0% |
| Hotel Magüipí | 765 | 4.70 | 69.66% | 12.36% | 17.98% | 57.25% |
| Hotel Torre Mar | 575 | 4.57 | 70.51% | 20.51% | 8.97% | 0% |
| Hotel Steven Buenaventura | 406 | 4.17 | 62.34% | 31.17% | 6.49% | 6.7% |
The sharpest contrast is not in the ratings but in the responses. Magüipí answers 57.25% of its reviews and also leads satisfaction with 4.70. Neither Cosmos Pacífico nor Torre Mar answered a single one in the entire period. Cordillera, the hotel with the most accumulated reviews, reached just 0.92%; it is also the only one where negative reviews exceed a third of the total.
How reliable these figures are
The automated classification was not taken at face value. A random sample of 3% of the corpus with text, 51 reviews, was read and labeled by hand. In 46 cases the human label matched the model's; 4 turned out to be mixed opinions where both still reached the same category; there was 1 real disagreement: a review that listed problems with the room but closed by praising the location, so the model went with the last sentence.
That leaves an agreement rate of 90.2%, the margin with which any figure published here should be read. The process is traceable end to end: the raw data, the clean data, and the consumption-ready data are kept in separate layers, following the CRISP-DM model, so any number in the table can be traced back to the review that produced it.
Bibliographic record
- Document type
- Professional degree thesis
- Institution
- Universidad de Manizales
- Faculty
- Sciences and Engineering
- Program
- Data Analytics Engineering
- Year
- 2026
- Author
- Andrés Felipe Duque Caicedo
- Thesis advisor
- Andrés Alberto Osorio Londoño, PhD
- Co-advisor
- Edgar Rafael Jiménez López, MSc
- License
- Creative Commons BY-NC-ND 4.0