Academic Libraries User Perception: A VADER-Enhanced Corpus of 7,605 Google Maps Reviews

The dataset comprises a curated corpus of 7,605 Google Maps reviews sourced from academic libraries worldwide, designed to assess user satisfaction and service perception. Data acquisition was conducted from March to July 2025 using the Outscraper extraction tool. Library records, including names, geographic locations, and Google Maps links and other fields, were systematically compiled to initiate data extraction. The dataset underwent pre-processing and cleaning, with duplicate entries and ratings without textual feedback removed. For linguistic consistency, all non-English reviews were machine-translated using the Google Translate API, resulting in an entirely English-language corpus suitable for advanced analysis. A noteworthy aspect of the dataset is the deployment of advanced Natural Language Processing (NLP) techniques to quantify user sentiment. The VADER (Valence Aware Dictionary and sEntiment Reasoner) model was employed for sentiment analysis. VADER generated a Sentiment Score Compound for each review, standardizing sentiment polarity on a continuous scale from -0.999 (most negative) to 0.999 (most positive), with zero representing neutrality. This computational strategy enables an assessment of sentiment intensity, surpassing the granularity of conventional 1-to-5 star ratings. All preprocessing, including tokenisation and data cleaning, was implemented in a Python-based Google Colab environment using the NLTK library. The dataset is further augmented with extensive metadata and derived metrics to facilitate multidimensional statistical inquiry. Principal variables comprise review text (original and processed), author history (review and rating counts), owner responses with timestamp data, and temporal attributes (year, month). Methodologically, the dataset is structured to support non-parametric statistical analyses. Given that descriptive statistics (mean, median, standard deviation) and the Shapiro-Wilk test confirms a non-normal distribution, the data then can be optimized for Spearman’s rank correlation coefficient analysis. This could enable examination of monotonic associations among key variables, including the relationship between review length and sentiment score, and divergence between star ratings and computed sentiment polarity.