September 30, 2026
What do readers say? Sentiment analysis applied to book reviews
What Do Readers Say? Sentiment Analysis Applied to Book Reviews
Caroline Cristine Reis Bueno; Jéssica Eloá Poletto
DOI: 10.22167/2675-6528-202602769
Article derived from a Final Course Work (TCC), with content based on the student’s original work and adapted to the editorial format of the E&S Magazine with the support of the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege for textual synthesis and organization.
Summary
Given the decline in the number of readers in Brazil, as pointed out by recent research, the aim was to understand the factors that attract readers and the feelings evoked by reading, in order to develop strategies to foster the habit. To this end, a sentiment analysis was performed on one thousand reviews of the ten most-read books on the social network Skoob, collected through web scraping techniques. The data were processed using Python libraries, including NLTK for preprocessing and the fork Léxico para Inferência Adaptada (LeIA) for sentiment analysis in Portuguese. An initial exploratory analysis revealed the predominance of words with positive connotations, such as ‘loved’ and ‘incredible’, and the absence of negative terms among the most frequent ones. The recurrence of words such as ‘movie’ and character names was also observed, indicating emotional attachment and the influence of film adaptations. The sentiment analysis indicated that nine out of the ten books presented a positive “compound”, suggesting that most of the works evoked positive emotions in readers. The high frequency of descriptive terms contributed to high neutral scores. It was concluded that attachment to characters and the influence of other media are relevant points for readers, and understanding these feelings can assist in choosing readings and promoting the habit.
Keywords: LelA; Social networks; Skoob; Text mining; Web scraping.
1. Introduction
Reading is a fundamental habit for individual and social development, but recent data indicate a worrying scenario in Brazil. According to the most recent edition of the “Retratos da Leitura no Brasil” (Portraits of Reading in Brazil) survey (2024), published by Instituto Pró Livro (IPL), more than half of the Brazilian population, precisely 53%, is classified as non-reader. This category includes individuals who have not read any book in full, or parts of a book, in the last three months, including religious or didactic works. The study reveals that only 27% of respondents read a book in full during the same period. Comparing with the previous survey in 2019, the average number of books read per person has halved, dropping from 4.92 to 2.07 in 2024, which represents an estimated decrease of 6.7 million readers in the last five years. This decline is observed across all demographic profiles, not being restricted to a specific gender, race, age, or education level. Furthermore, only 43% of readers claim to have read by their own volition, without the reading being demanded by school, college, or work. Reading frequency also decreases with increasing age group, being predominantly high among young people aged five to thirteen.
This downward trend is alarming, especially considering the multiple benefits that reading provides. Several academic studies, spanning areas such as languages, communication, and neuroscience, demonstrate that reading is beneficial for everyone, regardless of age or social status. Initially, the habit of reading expands vocabulary, improves writing, critical thinking, and creativity. In the long term, studies indicate that reading can delay the effects of neurological diseases such as Alzheimer’s and dementia (Pontifical Catholic University of Rio Grande do Sul [PUCRS], 2026), as it stimulates neuroplasticity, the brain’s ability to adapt and reorganize neural connections. Focus and memory, both short-term and long-term, are also skills enhanced by reading (Ribeiro, 2025). Additionally, the positive impacts of this habit extend to society, with evidence suggesting that reading increases empathy (Ribeiro, 2025).
Given the panorama of declining reading in Brazil and the proven benefits of this habit, it becomes crucial to understand what attracts existing readers to the books they consume. Understanding the sentiments aroused by reading and the points that capture readers’ attention is essential for developing effective strategies that can foster the reading habit among non-readers and, at the same time, maintain the engagement of current readers. In this context, Natural Language Processing (NLP) emerges as a relevant approach for analyzing unstructured data, such as social media texts (Liu, 2012). Text mining and sentiment analysis, in particular, allow for the extraction and interpretation of opinions, sentiments, and attitudes expressed textually (Becker and Tumitan, 2013; IBM, 2025b).
The social network Skoob, a Brazilian platform dedicated to readers with over ten million active users (Pancini, 2024), offers a vast repository of book reviews, where users share their opinions and ratings. This feature makes the platform a conducive environment for investigating readers’ sentiments and motivations. Understanding the factors that influence the reading experience and the emotions evoked by books is fundamental for developing initiatives that encourage the reading habit in Brazil. Thus, this study is justified by the need to deepen the knowledge about Brazilian readers’ perception of literary works. The objective of this study was to conduct a sentiment analysis on comments taken from the Skoob social network, considering the ten most read books by the platform’s users. With this, we sought to understand the sentiments evoked by reading and the aspects that capture readers’ attention.
2. Material and Methods
This study was characterized as a Natural Language Processing (NLP) research, with a quantitative and qualitative approach, focused on the analysis of unstructured data. The methodology employed machine learning techniques for the extraction and analysis of the content of the reviews, specifically using web scraping, text mining, and sentiment analysis, with the support of libraries from the Python programming language (Liu, 2012).
The unit of analysis consisted of one thousand book reviews, collected from the Brazilian social network Skoob, a platform dedicated to readers with over ten million active users (Pancini, 2024). Skoob was selected because it is a Brazilian social network, with public comments and the requirement that reviews must have at least one hundred characters for publication, which ensures more in-depth content for analysis. Global platforms, such as Goodreads, were disregarded due to the predominance of English comments and the difficulty in distinguishing reviews from Brazilian users.
For data collection, initially, the ten most read books on the Skoob platform were identified, according to the list released by the social network itself. Subsequently, the web scraping technique was applied to extract the reviews. Web scraping is a procedure that uses software to crawl web pages, imitating human navigation to extract and store information in an automated and precise manner (Lawson, 2015).
The extraction of reviews was performed using the Python libraries BeautifulSoup (Leonard Richardson Revisionn [LRR], 2025) and Selenium. BeautifulSoup was employed to navigate and search for information on web pages. As the Skoob website structure displays only ten reviews per page, the Selenium library was used in conjunction to automate the navigation through ten review pages for each book, ensuring the collection of the necessary amount of data.
The selection criterion for the reviews was the publication date, with the hundred most recent reviews being extracted for each of the ten selected books, totaling one thousand reviews. After collection, the data were subjected to a text mining process, which consists of transforming unstructured data, such as social media comments, into a structured format that allows for systematic analysis (IBM, 2025a).
The text preprocessing step was performed with the Natural Language Toolkit library (NLTK Project, 2025). The procedures included punctuation removal, conversion of all text to lowercase, elimination of stopwords (articles and prepositions), and the application of stemming and lemmatization (Alves, 2021). Lemmatization, in particular, considers the context to return the base form of the word, unlike stemming which can generate non-existent words in the language.
For sentiment analysis, the Lexicon for Adapted Inference (LeIA) was employed, a “fork” of the Valence Aware Dictionary and Sentiment Reasoner (VADER) library (Almeida, 2018; Aprendiz Artificial, 2024). VADER, originally developed for English, functions as a dictionary based on rules that assign semantic scores to words, classifying them as positive, negative, or neutral, and providing a “compound” metric that summarizes the overall sentiment of the text on a scale from -1 (very negative) to 1 (very positive).
The decision to use LeIA, adapted for Portuguese, was made to preserve the linguistic characteristics of the original reviews, avoiding losses of nuances that could occur with translation into English (Bizzocchi and Simões, 2025). During the initial tests, manual adjustments were made to the Lexicon, adding colloquial words such as ‘adoro’, ‘amo’, ‘bem bom’, and ‘bem ruim’ to the dictionary, in order to improve the accuracy of sentiment classification.
Besides sentiment analysis, Python libraries were used for the creation of data visualizations, such as histograms and word clouds, to aid in the identification of patterns. The Seaborn and Matplotlib libraries were employed to generate histograms, while the Wordcloud library was used for the creation of word clouds from the reviews extracted from Skoob. These tools allowed for an initial exploration of the data, focusing on word frequency and sentiment distribution.
3. Results and Discussion
The analysis of data collected on the social network Skoob revealed significant insights into the factors that attract readers and the sentiments evoked by reading, aligning with the objective of understanding the motivations and emotions that drive the habit of reading in the Brazilian context. The initial exploration of one thousand reviews of the ten most-read books on the platform allowed for the identification of language patterns and emotional connotation, providing a basis for more in-depth sentiment analysis. The findings indicate a predominance of positive evaluations and a strong attachment to specific elements of the works, such as characters and plot, in addition to the influence of other media.
In the exploratory analysis stage, the frequency of the most used words in the reviews demonstrated a clear inclination towards terms with positive connotation. Words such as ‘good’, ‘love’, ‘loved’, ‘incredible’, and ‘liked’ figured prominently among the fifty most repeated, reflecting a generally pleasant and satisfactory reading experience for users. It is noteworthy that, in contrast, no word with a negative connotation, such as ‘hated’, ‘horrible’, or ‘bad’, appeared among the most frequent terms. This absence suggests that, for the most popular books, the expression of discontent is less common or less intense in public reviews.
Beyond terms expressing direct feelings, word frequency analysis also highlighted the recurrence of descriptive vocabulary, such as ‘characters’, ‘story’, and ‘film’, along with proper character names. The presence of these words indicates that readers connect deeply with the narrative and its protagonists, developing an emotional attachment that manifests in their reviews. This observation is crucial for understanding what truly captures readers’ attention, suggesting that immersion in the fictional universe and identification with the characters are central aspects of the reading experience.
A particularly relevant finding was the strong correlation between the most read books and their adaptations for cinema or television. The research identified that nine out of the ten books on Skoob’s most read list have audiovisual adaptations. This phenomenon corroborates Marinho’s (2017) observation, who pointed out that, between 2010 and 2016, most of the best-selling books in Brazil were works that had been adapted for cinema. The author exemplifies this trend with the book ‘Me Before You’, which only reached the best-selling list in the year its film adaptation was released.
This intersection between literature and other media suggests that film adaptations act as a catalyst for reading, driving public interest in the original works. Marinho (2017) argues that films can be effective tools for sparking young people’s interest in reading, advocating for adaptations to be incorporated into school activities. In the context of this study, the high frequency of the word ‘film’ in reviews of the most read books reinforces the idea that visual media plays a significant role in readers’ decisions and engagement with literary works, creating a cycle of cultural consumption.
Moving on to sentiment analysis proper, which used the “fork” Lexicon for Adapted Inference (LeIA) of VADER, it was possible to quantify the emotional polarity of the reviews. LeIA assigns scores for positive, negative, and neutral sentiments, as well as a “compound” value that summarizes the overall sentiment of the text on a scale from -1 (very negative) to 1 (very positive). This metric allowed for a systematic evaluation of the predominant emotional tone in the thousand reviews collected, offering a detailed view of readers’ perception of each work.
The results of the sentiment analysis by book revealed that the vast majority of works evoked positive emotions in readers. Specifically, nine out of the ten most read books presented a positive “compound” value. The book ‘The Lightning Thief’ stood out with the highest average “compound”, reaching 0.53, accompanied by a low average negative score (0.08) and a neutral average of 0.73, with a positive average of 0.20. This indicates that the reading experience of this book was predominantly favorable, with readers expressing satisfaction and enthusiasm in their reviews.
The only exception to this positive trend was the book ‘Harry Potter and the Order of the Phoenix’, which registered a negative “compound” value of -0.09. This book also presented the lowest average positive score (0.12) among all the books analyzed, although its average neutral score was relatively high (0.73), and the average negative score was 0.14. This particularity suggests that, for this specific work, the reading experience may have been more complex or ambivalent for a portion of readers, or that the reviews contained more descriptions than direct positive emotional expressions.
The distribution of “compound” values for each book, visualized through histograms, reinforces these observations. For ‘The Lightning Thief’, the histogram showed a high concentration of reviews with “compound” values above 0.75, and a tiny amount of reviews with “compound” less than zero. This demonstrates the strong positive polarity of the evaluations for this work. In contrast, the histogram of ‘Harry Potter and the Order of the Phoenix’ exhibited a more dispersed distribution of “compound” values across the entire scale, with a frequency peak at -1.0, indicating that a significant portion of the reviews expressed negative sentiments.
The analysis of word clouds for specific books also provided valuable insights into the nature of the expressed sentiments. In the word cloud of ‘Harry Potter and the Order of the Phoenix’, an almost total absence of words denoting explicit sentiments was observed. Most of the most frequent terms referred to specific characters and the plot, such as ‘ministry’, ‘order’, and ‘phoenix’. The only words with sentimental connotation were ‘anger’ (in the twenty-second position), ‘good’ (in the thirty-seventh), and ‘liked’ (in the forty-fifth), which may justify the negative “compound” and the high neutral score for this book.
On the other hand, the word cloud for ‘The Lightning Thief’ revealed a significant presence of words expressing positive feelings, such as ‘incredible’, ‘loved it’, ‘love’, ‘liked it’, ‘wonderful’, and ‘good’, which shared space with terms related to characters and the plot. This difference in the composition of word clouds between books with a positive and negative “compound” illustrates how the language used by readers directly reflects their emotional experience with the work, whether it be enthusiasm or frustration, or even a more descriptive approach.
Upon analyzing the aggregated histogram of “compound” values for all one thousand reviews, a high concentration of ratings with “compound” between 0.75 and 1.0 was observed. This finding reinforces the conclusion that, for the most part, the most read books on the Skoob platform generate very positive reviews. Although some reviews presented low “compound” values, the prevalence of positive sentiments in the ten analyzed books was undeniable, indicating that the popularity of these works is intrinsically linked to the emotional satisfaction they provide to readers.
The high average neutral score observed in almost all books can be justified by the high occurrence of descriptive words in the reviews, which refer to character names, locations, or aspects of the plot. These words, although they do not carry an explicit emotional connotation, are essential for understanding the plot and the reading experience. The abundant presence of descriptive terms, without a clear emotional polarity, tends to raise the neutral score in sentiment analysis, even in reviews that contain positive elements.
It is important to consider the challenges inherent in applying sentiment analysis tools developed for English, such as VADER, to the Portuguese language. The grammatical complexity of Portuguese, with its particularities in pronouns, verb tenses, and inflected infinitives (Bizzocchi and Simões, 2025), presents an obstacle. Although LeIA has been adapted for Portuguese, the presence of slang, abbreviations, and colloquial expressions in reviews can still impact the accuracy of the analysis, contributing to some sentences with positive content being classified as neutral, as observed in the initial tests that required manual adjustments to the lexicon.
In summary, the research results indicate that readers of the most popular books on Skoob are attracted to narratives that evoke predominantly positive feelings, with a strong attachment to well-developed characters and plots. The influence of other media, especially film adaptations, plays a significant role in engagement with literary works. The predominance of descriptive words in reviews, although contributing to neutral scores, does not overshadow the strong positive emotional charge expressed by readers, which is manifested in terms like ‘loved it’ and ‘incredible’. These findings respond to the study’s objective, revealing that the emotional connection with the story and its characters, combined with the visibility provided by adaptations, are crucial points for fostering and maintaining the reading habit.
4. Conclusion
This study sought to understand the feelings evoked by reading and the factors that attract readers, through a sentiment analysis of reviews of the ten most-read books on the social network Skoob. It was found that the majority of reviews expressed predominantly positive sentiments, with a high frequency of terms such as ‘loved it’ and ‘incredible’, and a notable absence of negative words among the most used. A strong emotional attachment of readers to characters and plots was observed, evidenced by the recurrence of descriptive vocabulary and proper nouns. Additionally, the significant influence of film adaptations was identified, with nine of the ten most-read books having audiovisual versions, suggesting that other media act as catalysts for engagement with literature. Sentiment analysis, using the LeIA fork, confirmed that nine of the ten analyzed works generated a positive “compound”, indicating a mostly satisfactory and emotionally engaging reading experience for the platform’s users.
Understanding these sentiments and points of attraction offers a valuable contribution to the development of strategies aimed at fostering the reading habit in Brazil and maintaining the engagement of current readers. By understanding the emotional load that a work demands, potential readers can make more informed decisions when choosing their next readings, aligning their expectations with the content. However, the analysis revealed that the high neutral score in many reviews is justified by the predominance of descriptive words, which, although essential to the plot, do not carry explicit emotional connotation. The complexity of the Portuguese language and the presence of slang and abbreviations in the reviews represented a challenge for LeIA’s accuracy, even after adaptations, indicating the need for continuous lexical improvements for future analyses. It is suggested that platforms like Skoob could implement mechanisms to classify books based on the predominant sentiments in their reviews, aiding in the curation and recommendation of works.
Bibliographic References
Becker, K.; Tumitan, D. 2013. Introdução à mineração de opiniões: conceitos, aplicações e desafios. In: Simpósio Brasileiro de Banco de Dados, 2013, Recife, Pernambuco, Brasil. Anais… p. 27-52.
Internacional Business Machine. n.d. O que é análise de sentimento? Disponível em: <https://www.ibm.com/br-pt/think/topics/sentiment-analysis>. Acesso em: 14 set. 2025b.
IPL, 2024
Liu, B. 2012. Sentiment Analysis and Opinion Mining, 1ed. Morgan & Claypool Publishers, San Rafael, CA, Estados Unidos.
Pancini, L. 2024. Skeelo compra Skoob e mira crescer no setor de e-books. O desafio: conquistar 10 milhões de
PUCRS, 2026
Ribeiro, M. 2025. Por que ler faz bem para o cérebro? O hábito da leitura traz uma série de benefícios para a saúde do cérebro, como melhora da memória, criatividade e até da empatia. Disponível em: <https://drauziovarella.uol.com.br/neurologia/por-que-ler-faz-bem-para-o-cerebro/>. Acesso em: 20 ago. 2026.
Article originating from the Final Course Work of the Specialization in Data Science and Analytics of the MBA USP/Esalq
To learn more about the course, click here and access the MBX Academy platform