Como dominar a análise de sentimentos baseada em IA em 2025?

How do your customers really feel about your product or service? Happy customers help your business grow. Unhappy ones can hurt it. The good news? You can track how they feel using sentiment analysis. In fact, 83% of software companies using sentiment analysis see big improvements in customer satisfaction within a year. Want to improve too? Here’s your guide to sentiment analysis!

Resumo rápido:

– Sentiment analysis is a way to figure out how people felt when they wrote the specific text. It can be negative, positive, or neutral and include more specific emotions like anger and joy.

– Businesses use it to understand what customers think and feel about their products, campaigns, or company in general. It helps with better decisions, stronger branding, and faster reactions to feedback.

– It’s used in social media monitoring, customer support, marketing, PR, and even politics. Anywhere people share opinions, sentiment analysis can help.

– Most tools use machine learning to learn from lots of labeled examples. They scan text, look for patterns, and decide the mood or emotion behind it.

– It’s not perfect! AI can miss sarcasm, struggle with context, or misread mixed emotions. Also, it often can’t explain why it labeled something the way it did.

O que é análise de sentimento?

A análise de sentimento (também conhecida como mineração de opinião ou IA de emoção) é um método de análise de dados de texto para identificar sua intenção.

It includes the process of analyzing online text to determine the emotional tone it carries. It originally aimed to detect whether sentiment around a brand or topic is positivo, negativo ou neutro.

O objetivo é reconhecer e categorizar automaticamente as opiniões expressas no texto para determinar o sentimento geral.

Simply put, sentiment analysis determines how the author feels about a certain topic.

• O sentimento positivo pode ser expresso por meio de palavras como "bom", "ótimo", "maravilhoso" e "fantástico".
• O sentimento negativo pode ser expresso por meio de palavras como "ruim", "terrível", "ódio" e "nojento".

But it’s not so simple… There are different types of sentiment analysis.

Let’s briefly talk about each of them.

Types of sentiment analysis

Here are the main types of sentiment analysis. Each with its own way of digging into what’s behind people’s talk:

1. Fine-grained sentiment analysis
   This sentiment analysis model goes beyond just “positive” or “negative.” It breaks things down into levels like very positive, positive, neutral, negative, and very negative.
   It can be star ratings or scoring something from 0 to 100. Perfect for surveys and customer feedback.
2. Emotion detection
   Here, the sentiment analysis algorithms aim to detect actual emotions like joy, anger, sadness, or frustration instead of just classifying negative and positive sentiment.
   For example, “I can’t stop smiling after using this app” would be tagged as happy, not just positive.
3. Intent-based analysis
   This one tries to figure out the goal behind the message. If someone says, “I struggled a lot to set up my account” it may detect the intent of asking for help and not just expressing frustration.
   Super useful for customer service and market research.
4. Aspect-based sentiment analysis
   This method zooms in on specific parts of a product or service.
   If someone writes, “The app is awesome, but the notifications drive me crazy,” the system knows the bad review is about the notifications strategy, not the whole app.

What is sentiment analysis used for?

Sentiment analysis can be used across nearly every industry to track and understand public opinion.

It helps businesses, organizations, and even governments determine what people think, feel, and say about topics that matter to them.

Here are some of the most common sentiment analysis use cases:

01 Gerenciamento da reputação da marca

People talk about brands online all the time, and these conversations are full of valuable insights. Sentiment analysis tools help track what customers’ sentiment is toward your brand, product or campaign.

Tracking customer sentiment online helps PR teams, marketers, and brand managers monitor how their brands are perceived and make more tailored decisions to improve their image.

02 Customer feedback analysis

Customers don’t just leave feedback in survey responses anymore. They tweet, post, blog, and comment.

A sentiment analysis tool can automatically flag whether that feedback is good, bad, or somewhere in between.

For example, UBER used sentiment tracking to check if users liked the new version of their app – without waiting for a formal survey.

Smart, right?

03 Conducting market research

Want to know what people think about your industry, a new trend, or your competitor’s product?

Sentiment analysis is perfect for market research. It can tell you what people love, what they don’t, and what they’re hoping for.

04 Prevenção de crises

Sometimes, a few unavoidable news articles or online reviews can snowball into a full-blown Crise de relações públicas. Sentiment analysis tools help you catch those early warning signs.

With real-time alerts for negative mentions, teams can quickly address issues and calm the storm before it spirals.

It’s a lifesaver for reputation management, especially in industries where timing is everything.

05 Political and social research

Sentiment analysis isn’t just for business. It’s widely used in politics and public affairs.

For example, during elections, researchers and campaign teams use it to track public reactions to candidates, policies, or speeches.

In 2012, the Obama campaign reportedly used sentiment analysis to monitor responses to political messaging.

Today, it’s used to analyze everything from global news reactions to online movements and policy feedback.

Por que a análise de sentimentos é importante?

Graças à realização da análise de sentimentos, você poderá:

• Get a better understanding of your sentimento de marca beyond the survey responses
• Don’t waste time on unstructured data
• Obtenha insights que o ajudarão a melhorar seus produtos e serviços
• Automate customer service and make your business more responsive
• Reagir rapidamente ao sentimento negativo e reverter a situação
• Protect your brand’s reputation
• Mantenha seus clientes satisfeitos, colocando sempre os sentimentos deles em primeiro lugar

Como fazer a análise de sentimentos?

Performing accurate sentiment analysis without using advanced sentiment analysis tools can be difficult.

Por que isso acontece?

As Krzysztof Rajda highlighted, analysis based on a large volume of data is time-consuming.

Sure, you can try to research and analyze mentions about your business on your own, but it will take up a lot of your time and energy.

Furthermore, the risk of human error is quite significant in that case.

For example, Duolingo gained over 89k mentions just during the last 30 days.

Can you imagine analyzing each of them and judging whether it has a positive, negative, or neutral sentiment?

I honestly can’t.

And that’s why you should use ferramentas de análise de sentimentos.

They do all the counting for you!

I used Brand24 and quickly got results for Duolingo:

• 30k of those mentions are positive
• 17k are negative
• & the rest is neutral

It took me less than 3 minutes to discover this insight.

Here’s how:

I set up my social listening project, clicked on the filtro de sentimento, and got this overview:

But, advanced sentiment analysis tools can do much more!

Dê uma olhada:

Real-world sentiment analysis case study: Duolingo

Duolingo had some ups and downs recently due to the AI replacements within the company controversy.

So, let’s find out what exactly happened!

By analyzing sentiment from social media mentions, we can gain valuable insights into public perception of the Duolingo brand.

01 Overall sentiment

The overall sentiment around Duolingo remains positive, with:

• 34% positive mentions
• 19% negative mentions

02 Sentiment by categories

Duolingo’s sentiment looks very different across different social media platforms.

Most of the buzz happens on X (Twitter). There, people express their frustrations most.

On the other hand, Instagram has mostly good vibes.

Duolingo should definitely focus on addressing the negativity on X.

That’s why it’s so important to check how sentiment shapes in different places. It allows brands to better tailor their efforts and messaging.

03 Sentiment over time

In the last 3 months, Duolingo received almost 300k mentions. Over 93k are positive, 53k are negative, and the rest are neutral.

Looking at the sentiment chart, you see the rise of negative mentions around April.

It was also the time when more negative mentions occurred than positive ones.

After selecting April and May as a time range, I used the AI Events feature and discovered that fans didn’t like the AI replacements in the company:

Monitoring sentiment over time helps identify emerging trends in customer feedback and public opinion, allowing you to spot new shifts or patterns as they develop.

Como você pode ver, graças à análise de sentimentos, é possível monitorar facilmente as mudanças nas emoções dos clientes.

04 Sentiment by topic

Graças ao Análise de tópicos de IA feature, I discovered the most important and trending topics related to Duolingo:

And I can easily check what their sentiment is by clicking the “more stats” button.

Here’s what it looks like based on the most popular topic – Duolingo Memes.

• This topic has 27% of positive mentions and 21% of negative mentions.

• The most common emotions within the topic are admiration and disgust.

05 Sentiment of speaking authors

Now, let’s detect who is talking about Duolingo in a positive and negative way.

You can find this analysis in the “Influencer Analysis” Tab.

Here are the most positive people:

And here are the most negative people:

06 Context of a discussion

During the last 3 months, discussions around Duolingo were mostly around words like:

O que é uma pontuação de sentimento?

One of the means to assess sentiment is the sentiment score.

Pontuação de sentimento é um sistema de escala que reflete a profundidade emocional das emoções em um texto.

• It detects emotions and assigns them sentiment scores, for example, from 0 up to 10, from the most negative to the most positive sentiment.

How to calculate it?

There are various ways to calculate a sentiment score, but the most common method is to use a dictionary of negative, neutral, or positive words.

To calculate a sentiment score, various factors are taken into account, such as:

• The number and type of emotions expressed
• The strength of those emotions
• The context in which they are used.

Como funciona a análise de sentimentos?

The science behind the process is based on algorithms of natural language processing and machine learning to categorize textual data as positive, neutral, or negative.

Sentiment analysis technologies might use various types of algorithms.

What kind of?

01 Automatic

This type of algorithm is powered entirely by machine learning.

It learns from the data it receives, making machine learning a key building block of artificial intelligence.

Como?

1. The process begins with a dataset of texts that are labeled as positivo, negativoou neutro.
2. Using this labeled data, the algorithm starts to train itself.
3. Over time, it learns to recognize patterns and eventually becomes capable of analyzing new, unseen texts on its own.

That’s why the quality of the training data is so important!

If the algorithm hasn’t seen a certain type of example before, it may struggle to assess it correctly.

And the biggest perk of this method?

It can handle huge amounts of data... much more than a rule-based system ever could.

But, machine learning models often work like a black box. They make decisions, but it’s hard (or impossible) to explain why a specific text was labeled as positive or negative.

02 Rule-based

This algorithm is based on manually created lexicons that define positive and negative words.

• As regras podem ser definidas em relação a outros aspectos do texto, por exemplo, parte do discurso, sintaxe e muito mais.
• The algorithm then analyzes the amounts of positive and negative words to see which ones dominate.

This approach is easy to implement and transparent when it comes to the rules standing behind analyses.

But, as rule-based sentiment analysis relies on predefined rules and lexicons, it is less effective at handling complex language features such as negation and metaphors.

03 Hybrid

This one combines both of the algorithms mentioned above and seems to be the most effective solution.

It’s because it combines high accuracy provided by machine learning and stability from the rule-based, lexicon-based approach.

In advanced tools, these algorithms tend to be mixed.

So, I asked Krzysztof Rajda, Head of AI at Brand24:

Como funciona a análise de sentimentos na ferramenta Brand24?

Brand24’s sentiment analysis relies on a branch of AI known as machine learning by exposing a machine learning algorithm to a massive amount of carefully selected data.

It can automatically learn and improve from experience.

Our AI Team incorporated recent advantages of Pretrained Language Models (PLM), a technique also used by Big Tech companies (like Google, Microsoft, Facebook, or Baidu).

Por quê?

It results in the machine’s capabilities being as close as possible to the real understanding of the text with the current state of scientific knowledge. 

What’s more, the usage of multilingual PLM allows us to perform sentiment analysis in over 100 languages of the world!

If you’d like to learn more about this process, we recently contributed to the science with our work about multilingual sentiment analysis.

It was presented at one of the most notable and prestigious conferências científicas.

Desafios da análise de sentimentos

Embora existam muitos benefícios da análise de sentimentos, you need to be aware of its challenges too.

Due to language complexity, sentiment analysis has to face at least a couple of issues.

In some cases, it gets difficult to assign a sentiment classification to a phrase.

That’s where natural language processing-based sentiment analysis comes in handy, as the algorithm attempts to mimic regular human language.

So, what are the most common sentiment analysis challenges?

01 Contrastive conjunction

One problem a sentiment analysis system has to face is contrastive conjunctions.

They happen when one piece of writing (a sentence) consists of two contradictory words (both positive and negative).

• Exemplo de frase: "O tempo estava péssimo, mas a caminhada foi incrível!"

02 Named-entity recognition

Outro grande problema enfrentado pelos algoritmos é o reconhecimento de entidades nomeadas. As palavras no contexto têm significados diferentes.

• "Everest" se refere à montanha ou ao filme?

03 Anaphora resolution

Also known as pronoun resolution, describes the problem of references within a sentence: what a pronoun, or a noun, refers to.

• Example sentence: “We went to the theater and had dinner. It was awful.”

04 Sarcasm

Sentiment analysis tools struggle a lot to detect sarcasm. They can even mark it in the opposite way.

• Exemplo de frase: "Estou tão feliz que o avião está atrasado".

05 The Internet

It just so happens that any language used online takes its own form.

The economy of language and the Internet as a medium result in poor spelling, abbreviations, acronyms, lack of capitalization, and poor grammar.

Analyzing such pieces of writing may cause problems for sentiment analysis algorithms.

And finally, there’s also a data struggle that tools face:

How industries use sentiment analysis tools?

Here are some examples of how sentiment analysis is used in different industries:

1. Retailers use sentiment analysis to:

• Improve customer satisfaction
• Optimize product development
• Gerenciar a reputação da marca

Real-time analysis of social media and customer reviews enables them to track shifting preferences, personalize promotions, and reduce negative feedback.

According to Journal of Big Data, it can help improve retention and up to 95% increases in profit.

2. Insurance companies apply sentiment analysis to:

• Detect dissatisfaction in customer communications
• Flag risky or fraudulent claims

By analyzing tone and emotion in policyholder interactions, insurers can improve customer experience and streamline claims processing in increasingly digitized workflows.

3. Manufacturers use sentiment analysis to improve:

• Product quality
• Supply chain responsiveness
• Predictive maintenance.

In fact, over 55% of manufacturers now use AI tools for these purposes, reflecting a trend toward data-driven quality control (source: Deloitte).

4. Media and entertainment organizations apply sentiment analysis to:

• Personalize content delivery
• Track viewer engagement
• Optimize marketing campaigns

As the OTT market grows rapidly, studios and platforms use sentiment tools to include audience reactions across social platforms in their plans.

5. Finance companies use sentiment analytics for:

• Decision-making aligned with public and investor sentiment
• Enhanced risk assessment and trading strategy development

Analyzing sentiment from financial news, earnings calls, and social media enables institutions to forecast market behavior and adjust their portfolios accordingly.

6. Telecom providers use sentiment analysis to:

• Identify churn risks
• Proactively intervene with service enhancements or targeted retention offers

Monitoring customer interactions across channels helps telecoms detect dissatisfaction early and implement corrective measures to improve loyalty.

7. Political organizations use sentiment analysis to:

• Analyze public sentiment on policies and services
• Guide decision-making during crises
• Improve communication strategies

By evaluating voter and citizen sentiment in real time (especially during elections or emergencies), government can tailor their messaging, build trust, and respond more effectively to public concerns.

8. Healthcare use sentiment analysis to:

• Analyze patient feedback
• Uncover service gaps
• Enhance satisfaction and care quality

Hospitals can use sentiment tools to monitor emotional tone in patient reviews and survey responses, helping improve doctor-patient communication, improve treatments, and increase overall patient satisfaction.

9. Sports organizations use sentiment analysis to:

• Understand fan perception
• Gauge campaign effectiveness
• Reduce manual survey costs

PERGUNTAS FREQUENTES

What is NLP sentiment analysis?

Sentiment analysis is built on natural language processing (NLP) in general.

It allows computers to process and make sense of human language and unstructured data.

Today’s NLP tools are far more advanced, making it possible to detect emotions and opinions with greater accuracy and depth.

Common NLP methods used in sentiment analysis include:

• Lexicon-based techniques that use sentiment word lists to assign polarity
• Machine learning models trained on labeled data to classify text by sentiment
• Deep learning frameworks like BERT and GPT that understand context and subtle tone
• Aspect-based sentiment analysis, which focuses on specific parts of a product or service to determine targeted sentiment

Here’s an expert explanation by Krzysztof Rajda, diretor de IA da Brand24:

Natural Language Processing (NLP), a part of AI that deals with textual data, is an essential technique in the modern world for discovering the unknown from monitoramento de mídia results.

One of the most useful NLP tasks is sentiment analysis – a method for the automatic detection of emotions behind the text.

Sentiment can be analyzed at different levels – from identifying positive, negative, or neutral opinions, quantifying the level of positivity or negativity, to even opinion mining and identifying the fine-grained emotion behind the opinion (e.g., happiness, anger, sadness, etc.).

What is the difference between sentiment analysis and emotion detection?

Sentiment analysis and emotion detection are related but distinct tasks in natural language processing (NLP).

1. Análise de sentimento focuses on determining the overall attitude or opinion expressed in a text
2. Emotion detection goes deeper by identifying specific emotional states such as alegria, raiva, tristeza, medoou surprise.

While a sentiment analysis model provides a broad sense of polarity, emotion detection captures the nuanced emotional tone behind the words.

For example, the sentence “I can’t believe they forgot my order again” would be labeled as negative in sentiment analysis, but emotion detection might identify it as expressing frustration ou raiva.

Can ChatGPT do sentiment analysis?

Yes, ChatGPT can perform basic sentiment analysis.

It can read text and suggest whether the tone is positive, negative, or neutral. It may also detect some emotions like joy, anger, or sadness.

However, there are a few drawbacks. ChatGPT doesn’t consistently follow strict sentiment labels like specialized sentiment analysis tools do.

Plus, its analysis isn’t automated or scalable – you’d need to feed in text manually or build a custom workflow to analyze large volumes of data.

Então, o que está esperando? Comece a se aprofundar nos dados!