How to use modern language models for enhanced sentiment analysis

The mood analysis promised to discover the secret feedback information, but outdated methods left us with shallow insights and too much simplified charts. With the increase in advanced language models, it is time to get rid of these restrictions and revolutionize the way we understand the feelings of customers – revealing the context, emotions and intentions that drive opinions.

From Italian flags to the shitty lines: the story of the analysis of feelings

Sentiment analysis has been around for a long time. Most companies use a version to analyze large quantities of text – either from social networks posts, answers to surveys or comments on the website.

Usually each clip of the text is categorized as “positive”, “negative” or “neutral”. The results are often shown using a horizontal chart with green, white and red stripes. They represent the ratios of positive, neutral and negative comments, which leads to a classic display of feelings of the “Italian flag”. There is a great chance that your organization has already used similar screens. They can be interesting, but often have a limited value.

Another usual approach is drawing percentages of a positive and negative mood over time, resulting in two curved lines that, although visually attractive, offer a little insight that can be taken. Early, some companies even tried to create “NPS feelings” by taking away the percentage of negative comments from the positive – another shirting line and another measure of limited use.

Despite these defects, the sentiment analysis was not completely ineffective. Its real potential lies in overcoming simple positive and negative percentages, which are often too variable to make sense.

The categories of feelings can serve as a starting point for a deeper analysis. For example:

• What topics are people discussing with positive feelings?
• How are they different from the negative topics mentioned?

Visualization techniques such as a butterfly chart, which counter the positive mentioned topics on the one hand and negatively on the other, can highlight useful insights for improving user experience.

This probably seems familiar to you. For over 15 years, the promise of drawing valuable marketing insights from open comments has been tempting, but its practical value remained limited.

The good news is that recent progress in linguistic models allows you to reconsider the analysis of feelings in new and exciting ways. With these modern tools, we can discover what triggers customer perception of our products and services.

Sentimena’s traditional analysis restrictions

Before exploring the new opportunities offered by modern language models, it is important to understand the limitations of traditional feelings analysis. Without this understanding, we risk repeating the same problems with new technology. Three primary problems limited the effectiveness of conventional mood analysis:

Lack of context

Traditional feelings analysis usually works with clippings of the text, often not considering the context required for the exact interpretation of meaning. For example, take a comment: “You are amazing.”

Without context, the feeling can be positive or negative. The feeling is probably negative if a person has rated their satisfaction with your service with 1/5. If you were rated 5/5, it’s probably positive. The context is critical, but the traditional mood analysis rarely includes it.

Ambiguity in a neutral category

One of the reasons for changing the percentages of positive and negative sentiment is a neutral category, which often combines two very different groups:

• Honestly neutral comments.
• Comments where feelings cannot categorize.

These groups should be treated separately, but traditional systems are no different. Many service providers avoid solving this problem because of its complexity, leaving this ambiguity unresolved.

Excessive simplification of human expression

Categorizing text as a positive or negative too much simplifies the complexity of human language. One comment can serve for multiple purposes or express contradictory emotions, making a binary approach insufficient. This restriction forces frequent relying on a neutral category, which further reduces the accuracy and value of the analysis.

Understanding these limitations is crucial to building better systems that avoid traps of traditional mood analysis.

Kopaj deeper: How to increase market research and gain insights into clients using artificial intelligence

How modern language models change the game

Modern language models can revolutionize a sense of feeling, but only if we avoid repeating the disadvantages of traditional methods. Creating a system for analysis of feelings using a modern large linguistic model (LLM) is surprisingly simple, but it still encounters the same limitations.

For example, consider the following inquiry:

• “Please categorize the following statement as positive, negative or neutral. Give me a category of just one word: ‘You are amazing.’

When using Gemini 1.5, the answer is:

Although LLM triggers this Sentiment analysis system, its effect is limited as traditional approaches. It is simply easier to implement.

In order to really use the potential, we must recognize the limitations of traditional mood analysis and strive for more advanced solutions. Here are some ways in which current models can solve these challenges:

• Include context: Mood analysis should include contextual information, such as satisfaction, previous interaction or other relevant data, instead of an analysis of isolated text clips.
• Adopt a sophisticated categorization scheme: Translate the basic positive, negative and neutral markings to include categories such as frustration, admiration, gratitude or sarcasm – adapted to what is most important for improving user experience.
• Focus on the purpose, not just emotions: Analyze the intent or mood behind the statements, whether descriptive, sarcastic, insecure or informative. Understanding these shades can lead to deeper insights.

A new wave of sentiment analysis appears, one that transcends the outdated visualization of the “Italian flag”. By effective use of modern tools, we can ensure that our mood analysis brings effective, business relevant insights.

Kopaj deeper: From feelings to empathy: understanding how customers feel