Distinctive Analytics

What is sentiment analysis?

Sentiment analysis can determine the polarity of emotions in given sentences. We can classify them into certain ranges 

Positive, neutral, negative 

We can extend these ranges for more precision

       Strongly positive, positive, neutral, negative, strongly negative

We can increase the ranges as per our applications.

We can employ machine learning algorithms and statistical methods with NLP to perform sentiment analysis and prediction.

Types of Sentimental Analysis

• Fine-grained sentiment
• Multilingual
• Intent analysis
• Aspect based
• Emotion detection

1. Fine-grained sentiment analysis

It’s as Sherlock Holmes says, “The small details are always most important.”

Fine-grained analysis works on the sub-sentence level. It works to identify the emotion of the sentence towards a subject or an object.

In this analysis, we can extract

What topic is being talked about?

Who’s talking about it? And

How do they feel about it?

The sentences also help us understand comparison and the intensity of emotions.

Like

This comment

“AVATAR 2009 was the film of the decade, whereas AVATAR Way of Water doesn’t even qualify for the movie of the year.”

In this comment, avatar movies are being compared. One movie is being praised while the other is being critiqued. There are two distinct sentiments in this review. Fine-grained sentiment analysis is able to extract sentiments on different topics from the sentences. While other methods might would have given mixed results from this comment.

2. Multilingual sentiment analysis

When people review a product online or give feedback. They are more expressive when they are using their native language. Rather than only looking for English reviews and comments, to collect as much data as possible, we can expand to local languages.

Multilingual sentiment analysis works across multiple languages. It is very handy as you can work on local languages to understand emotions in proximity to where you provide or supply your service or product.

If you are serving a particular demographic, you may as well understand their language. Through the means of multilingual sentiment analysis, that is.

There are many challenges in these fields too; as the language changes, so do the common phrases, idioms, and words with multiple intents and meanings. Work is ongoing in this field, and it’s always evolving into a better version.

Python libraries like NLTK and Spacy have strived to put more languages under their belt for processing. Spacy alone supports more than 50 languages by now.

3. Intent analysis

Intent analysis deduces a person’s motivation from their writing. You can apply intent analysis in many fields where you need to know the intent behind a written text.

Consider that you have thousands of comments on a newly launched product. Now you have to evaluate the impression it made on people.

You need to understand if people are curious about features.

Are they curious about some features they don’t understand?

How do they feel about the product?

Will they refer it to others?

If you understand these intentions in time, you will be able to take the necessary action, whether it be out of desire, curiosity, or complaint. It’s best to understand and address important comments. So they can bring more attention to your product and also have your users satisfied with both the product and your attention.

4. Aspect-Based Sentiment Analysis

Aspect-based sentiment analysis focuses on a person’s reaction to a specific topic in a matter. ABSA analyses users’ text for a particular aspect of the service or product.

For that matter, take the Uber cab service. When you complete a cab ride. You get feedback forms on multiple matters, like cleanliness, service, navigation, value for money, time, and conversations.

These are the aspects Uber wants to maintain the quality of its service. Uber may have streamlined the process of ABSA by directly asking about the aspects themselves. Normally, we have to scrape or gather feedback from users on the internet and find out different aspects they write about.

5. Emotion Detection

Emotion detection recognizes the six fundamental emotions of a person: fear, anger, joy, sadness, contempt, disgust, and surprise.

Emotion detection is also known as emotion recognition. But this can be confusing, as it could be interpreted as sentiment analysis, as they have similar meanings.

The basic difference between plain sentiment analysis and emotion detection is finding polarity.

Sentiment recognizes the polarity of emotions as positive – neutral – and negative.

Meanwhile, emotion detection is for the full spectrum of human emotions.

Methods of sentiment analysis

• Lexicon based method
• Machine Learning method
• Hybrid methods

1. Lexicon based Sentiment analysis

A lexicon-based approach to sentiment analysis uses pre-trained models as its basis, where rules called lexicons are used. These rules classify words into positive and negative ones, and they carry their intensity level as well.

Negation terms such as “can’t,” “no,” “don’t,” and “shouldn’t” are secondary. The first are lexicon words.

Sentiment Analysis with Machine Learning

1. Naive Bayes

Bayes has been a supervised ML algorithm since Naive. It needs labelled data to train first. Naive Bayes uses probability classification. Naive Bayes can classify between polar data of type negative and positive. Give some labelled data to feed first.

2. Logistic Regression

Logistic regression often uses binary classification. The state of the result of binary classification remains from 0 to 1.

3. Support Vector machine

SVM can be used for both classification and regression. SVM uses hyperplanes and classifies data into vectors in n-dimensions. Using SVM, texts are separated into different dimensions, and new data is classified according to their similarity to these words.

Sentiment Analysis with Deep Learning

1. CNN

As we know, CNNs are mostly used for image processing. But we can also use it for NLP, as images are converted into a one-dimensional array before they are fed to a CNN. Similarly, a sentence is already a one-dimensional array. So after a few tweaks, we can use it for sentiment analysis.

2. RNN

RNNs retain a temporal memory of trained data. RNNs can be used to understand sentences. We can use that to analyse sentiment from RNN.

RNNs have a problem with vanishing gradients. Throughout the training, the old inputs become less memorable and disappear over time.

LSTM are an extension of RNN

LSTM give RNN the ability to remember or forget new information and assign weights to information that affects the output.