Social media today has become a critical component in our lives. We learn, create, express, influence, and even get influenced on social media platforms. Political campaigns, marketing campaigns, businesses, and prediction based decision making runs on sentiment analysis.

Sentiments analysis, a data mining technique that is used to measure and understand people's opinion and inclination through NLP. Computational linguistics and text analysis are used to analyze information from the web, social media, and other such sources. 

This data quantifies people's emotions, sentiments, beliefs, or viewpoints; hence is also called opinion mining. This uses NLP to observe people's mood at large in society through the data from social networking on the web. 

The key idea is to use NLP techniques, especially semantics and word sense disambiguation, for more accurate opinion mining. Word sense disambiguation in NLP is the ability to determine which meaning of the word is activated by the use of the word in a particular context. 

Social networking sites use NLP techniques such as speech tagging and relationship searching to identify sentence components such as subjects, verbs, and objects. These entities are then analyzed to establish an underlying relationship to understand whether the sentiments attached are negative or positive. 

Data in the form of multimedia, text, and images are what is raw data for NLP based sentiment analysis. These data sets, when analyzed, can convey the tone or attitude of the poster. This mined text information is subjected to ensemble classification for analysis that involves syncing of various independent classifiers, and then they are further classified under different tones. 

It has also been demonstrated that ensemble classification is an improved version of traditional machine learning classifiers. Various ML algorithms such as SVM (Support Vector Machines), Naive Bayes, and MaxEntropy work on data classification.

One major concept at the backend is word embedding, which is a representation of words in the form of vectors. Each word is mapped to one vector and the vector values are learned in a way that resembles an artificial neural network. Every word vector then becomes a row of real numbers where each number is a dimension of the word's meaning. So it leads to semantically similar words having similar vectors ie, synonyms will have equal or close vectors. 

The input data for algorithms operating around word embedding is a large corpus of text which is converted into vector spaces. Word embeddings are one of the most successful AI applications of unsupervised learning. The most popular word embedding algorithms are Google's Word2Vec, Stanford's GloVe, or Facebook's FastText.

In the Indian context, Prizmatics, headed by Ajay Piwhal is working on the same lines to identify and categorize emotions from a vast amount of data. They are implementing NLP and Deep Learning in the field of text analytics and sentiment analysis in India to mount an accuracy greater than 95% for clients of various industries.

Sentiment analysis helps you understand or reach a crux of the matter, predict a crisis, a wave, reach out to unhappy customers, or help run a marketing/political campaign. As otherwise, it would be impossible to scan through all the posts or all the available texts on social media to reach to a conclusion or get the underlying idea. This analysis helps convert unstructured text into structured data using NLP and open source tools.  

Today, many businesses use hashtags, Twitter data, Facebook drives today to understand opinions and focus in the right direction. 

Challenges involved in Sentiment Analysis:

• Anaphora resolution - This issue arises owing to confusions around dangling modifiers or mismatching references.
• Named entity recognition - recognising and classifying entities texts into pre-defined categories such as name, place or other such other nouns.
• Parsing - The segregation of sentence into subject or object and other parts of speech such as adjective or verb or pronoun is the challenge that requires more accuracy.
•  Rhetorical modes - Detecting sarcasm, irony, comedy, etc also a difficult task that NLP algorithms need to work upon.
• Social media website - Casual writing styles on social media platforms leads to a challenge in accurate assessment of the data in hand.
• Visual sentiment analysis - The data present out there is mostly a mix of visual and textual information pose as a challenge to sentiment analysis.

These challenges pose as the right direction in which improvement is required. Brand monitoring, customer service, and market research are already utilising text analytics. Moreover, sentiment analysis is set to make its mark in the areas of political science, sociology, psychology; flame detection, identifying child-suitability of videos, bias identification in news sources and many more. So every time you post something on social media, you know that at the backend, this is reaching a huge ocean of data that can be studied to understand the deeper insights and emotions.

Photo by ThisIsEngineering from Pexels