Try our custom LLM Masker


Collect actionable information about positive or negative opinions on social networks, review sites, surveys and customer service applications

Do you need to find out what the public thinks in any language?

Talk to an expert

Why use a Sentiment Analysis tool?

The exponential increase in User-Generated Content, such as product reviews by users who give their opinions, as well as instant comments on social networks and blog posts, call for tools that can provide a solution to quickly find out the public's opinions on a wide range of topics, in any language.


Pangeanic's Sentiment Analysis tool is customizable, powerful and efficient technology for automatically extracting positive or negative opinions from written text. It can be tuned to detect specific emotions (strong dislike or liking, fear, anger, disgust, etc.) from unstructured textual information. It is an API prepared for immediate analysis within sentences (fragments) or for batch processing of texts and documents. Pangeanic's Sentiment Analysis tool can also be used for other purposes, such as document opinion classification and review rating prediction tasks.

How does the Sentiment Analysis tool work?

There are two main approaches to Sentiment Analysis: traditional "lexicon-based", and the new "learning approach".

The traditional approach to Sentiment Analysis is based on a lexicon. Using this approach, the semantic orientation of the words in a text is calculated by obtaining the polarities of the words from a predefined lexicon. The supervised learning approach uses strong neural Machine Learning techniques to create a specific model from a large corpus of documents. This data varies from one client to another and may include relevance screening (a prior consultation to verify if a document belongs to the group and should be analyzed).


A set of samples of positive and negative opinions and variations of those opinions make up the training data from which the model is built. Our Machine Learning techniques achieve more than 80% accuracy from the very beginning. Customization and human-created linguistic filters in each case add to the expected quality to obtain an accuracy rate of more than 90%. The difficulty of Sentiment Analysis is due to the structure of opinions, often field-dependent and sometimes context-sensitive.

The unique hybrid approach of Pangeanic's Sentiment Analysis tool is based on the use of both linguistic and statistical information, as well as a set of complex language-dependent semantic rules. This, together with the use of deep neural networks to classify irony and seemingly negative or positive results, provides a rapid, unique opinion classification solution.


Approaches for successfully processing Sentiment Analysis:

Our linguistic experts have created customized ontologies for the most significant and widely used data fields (hotels, restaurants, appliances, etc.). These used to be the basis of Sentiment Analysis, but now feed deep neural networks for which data has been previously labeled. Opinions can be grouped in different ways according to customer preferences.

In the case below, orange represents a negative opinion, yellow, a neutral one and blue, a positive one. These are real samples from the combination of our hybrid techniques:


Negative: the play was short
Positive: I love how short the journey was

Neural detection (context)
Positive: Bailey made sure I didn't go short. He's quite the artist.

Negatives/intensifiers (neural detection)
Positive: This cheeseburger is not bad at all
Negative: Service too fast for the price

Field-dependent opinions
Positive: Small. Lightweight. Portable. 16Mb. Thanks, Dad!
Neutral: Not all funds are created equal.