Do computers understand emotion? When a news story states “Tesla continues to crush the competition,” does a computer understand the true meaning, or does it get hung up on negative words like “crush?” The answer to this until recently would be no – most sentiment analysis models cannot recognize such linguistic nuances and would confidently classify this statement as neutral or even negative, giving erroneous data to the company.

That’s because the majority of sentiment models on the market today rely on relatively crude methods that analyze words in a sentence in isolation and average them together to get a final tally that categorizes it as either a positive, neutral, or negative statement. Unfortunately, this means that they often miss the bigger picture and fail to take into account the nuances of language. So, when a model picks up words like “crush” in a news story, it declares the sentence as negative when the sentiment behind it is obviously positive.

Signal’s new model uses target-based sentiment that understands the target of the statement and the sentiment of the author, delivering a 20% improvement in accuracy over the previous AI model.

It’s clear that traditional models simply labeling words as good and bad cannot quite capture the complexities of human language. As artificial intelligence technology evolves at a rapid pace, there is an urgent need for an emotionally intelligent sentiment model. To help solve this challenge, the Data Science team at Signal AI began developing a new entity-based sentiment analysis model that could analyze an entire section of text taking context into account. Signal’s new model uses target-based sentiment that understands the target of the statement and the sentiment of the author, delivering a 20% improvement in accuracy over the previous AI model. Since its launch, Signal AI’s cutting-edge sentiment model has outperformed the other models and shown significant improvements in accuracy.

A Giant Leap in Data Precision

Accuracy is the name of the game when it comes to sentiment analysis. But because most sentiment analysis models today can only consider individual words, they struggle not only to understand the context of a text, but also to identify the subject, or “entity” that the organization is trying to measure. This often results in inaccurate data. For the countless organizations around the world that rely on sentiment analysis to quantify the impact of media perception on their brand’s reputation, this can quickly prove to be an untenable situation. When thousands of mentions are misclassified by a sentiment analysis tool, comms teams not only waste hours tediously correcting analyses but, more importantly, miss key data points that inform areas of improvement and risk.

Signal AI’s cutting-edge sentiment analysis engine, powered by the same type of transformer technology underpinning ChatGPT, offers an alternative solution with its entity-based model that breaks down a text to look for patterns and nuances to discern the sentiment of the author. This results in much more accurate analysis, allowing companies to make data-driven decisions and adjust their marketing strategies in real-time. It also means less manual work and more time to work strategically. The new model makes sense of the avalanche of data that organizations face today, giving them more control over risk surfacing, customer satisfaction, and brand management.

For example, consider the passage below:

Previous models would have classified this as a neutral statement due to the lack of overtly negative words.

Now let’s look at how Signal’s new model would read the same passage:

The new entity-based sentiment model clearly recognizes this statement as a negative reference toward both McKinsey and French president Emmanuel Macron.

The Big Picture of Sentiment Analysis

The promise of accurate sentiment analysis hinges on a more granular approach that focuses on nuanced context. Traditionally, sentiment analysis relied on rudimentary methods such as individual words analysis, which resulted in inconsistent and incomplete data analysis. This is akin to expecting a model to tell us what the full picture is when it is only able to analyze singular pixels rather than the shapes they form. With the advancements in AI technology, the new entity-based analysis model provides more accurate and relevant data analysis.

As our lives become more digitized, sentiment analysis continues to be one of the most valuable tools in a company’s arsenal in assessing its reputation as well as foreseeing opportunities and risks. While sentiment analysis is still subjective to some extent, the new model represents a breakthrough in harnessing the power of machine learning to understand human communication. For companies seeking to stay ahead of today’s shifting business landscape, this model is a game-changer, giving companies an unprecedented level of insight not possible before.