SENTIMENT ANALYSIS

: A text based approach

Presented By :
Om Satpathy
2002040030
CSE
OVERVIEW
• Introduction
• Text based vs. Image Based
• Types of Sentiment Analyses
• Application and uses
• Advantages
• Step by Step process of SA
• Challenges
• Case study
• Thank you !
 Introduction
• Identification of sentiments from digital data.
• Digital data can be texts or images.

• Sentiment analysis is powered by natural language processing

(NLP) and machine learning (ML) algorithms. These
artificially intelligent bots are trained on millions of pieces of
text to detect if a message is positive, negative, or neutral.
• Text based sentiment analysis is an automated process that
determines the emotional tone behind a message.
> Text based Sentiment Analysis
> Image based Sentiment Analysis
Types of Sentiment Analysis
Emotion Detection
 Aspect Based Sentiment Analysis
• Also known as fine-grained opinion mining.
• It is the task of determining the sentiment of a text with respect
to a specific aspect.
• Traditional Sentiment Analysis vs. ABSA

Traditional Sentiment Analysis methods treat a text as a whole and assign it to a

single sentiment label(e.g., positive, negative or neutral).
While ABSA(Aspect Based Sentiment Analysis) refers to identifying the sentiment
of a text with respect to a specific aspect.

Example : A restaurant review – Overall review of restaurant is positive, except

the service. In this case, it would be useful to know “service” is negative, even
though overall restaurant experience is positive.
Example of Aspect Based Sentiment Analysis
 Fine grained Sentiment Analysis
• In most cases, Sentiment Analysis models are just programmed for binary
classification(positive or negative sentiment).

• Fine grained sentiment analysis classifies the text into 5 different

classes.
• This method is applied on a sub-sentence level and it is meant to
identify a target(topic) of a sentiment. A sentence is broken into
phrases or clauses, and each part is analyzed with connection to
others.
Examples of Fine Grained Sentiment Analysis
We can identify who talks about a product and what exactly a person talks
about in their feedback. In addition, it helps understand why a writer
evaluates it in a certain way.

Example : “SAMSUNG IS WAY BETTER THAN IPHONE.” Such

comparative expressions can be properly evaluated using fine grained SA as
compared to coarse-grained SA.

Example : “A TOUCHPAD ON MY LAPTOP STOPPED WORKING

AFTER 4 MONTHS OF USE.” It can also identify which aspect of a
product the user is talking about. This way, we know exactly what must be
improved or reconsidered.
 Intent based Sentiment Analysis
• Identifies the intent expressed in the text.
• an intent classifier automatically analyzes texts and categorizes them into
intents such as Purchase, Downgrade, Unsubscribe, and Demo Request.
 Examples of Intent Based Sentiment Analysis

• For analyzing customer emails, we might choose tags like : Interested,

Need Information, Unsubscribe, Wrong Person, Email Bounce, Autoreply,
etc.
• Example : “I TRIED TO MAKE A PURCHASE THROUGH THE SITE
BUT I DON’T KNOW WHERE TO START, COULD YOU HELP ME
OUT?” This example can be tagged as INTERESTED.
• To take intent detection one step further, you can combine it with text
extraction to identify specific data from text, such as locations, dates,
company names, etc., that are related to a user's intent.
• Example : “I WANT TO BOOK A FLIGHT FROM NEW YORK TO LAS
VEGAS, BUT MY CARD HAS BEEN DECLINED.”
o Text extractor – “Las Vegas” and “New York”.
o Intent classifier – “Book a flight”
 Applications and Uses
• Social media monitoring.

• Brand monitoring

• Voice of customer (VoC)

• Customer service

• Workforce analytics and voice of employee

• Product analytics

• Market research and analysis

 Advantages
• Scalability: Sentiment analysis allows to process data at scale in a efficient
and cost-effective way.

• Real-time analysis: A sentiment analysis system can help you immediately

identify these kinds of situations and take action.

• Consistent criteria: By using a centralized sentiment analysis system,

companies can apply the same criteria to all of their data. This helps to
reduce errors and improve data consistency.
SA – Step by Step Process
Step 1 : Tokenization
Tokenization Types
 Step 2 : Cleaning the data

• Remove numbers.
• Stemming or
lemmatization
• Part of speech tagging
• Remove punctuation
• Lowercase
 Step 3 : Removing the stop words

The idea is to remove all those words which appear commonly across all
the documents in the corpus e.g. articles and pronouns.
Most of the time, stop word list for the given language is a well hand-
curated list of words.
Step 4 : Classification
Step 5 : Apply classification algorithm
Step 6 : Calculation
Summarizing everything !
 Challenges
• Negation detection error : Example – “It was not unpleasant.” is a positive
word but can be classified by an algorithm as negative if not properly trained.

• Multilingual data : When multiple languages are used in a sentence. Example –

Instagram “see translation” feature.

• Potential biases in model training : Examples – Suppose a model labels “I can

be a sensitive person.” as negative and “I can be very ambitious.” as positive.

• Context dependent errors

o Sarcasm : Example – “I am so glad that the product arrived in one piece.”
Most S.A. models classify this as positive whereas in real life, this sentence
may be related to customer dissatisfaction about the delivery.
o Polysemy : When words have more than one meaning. Example – “Head
of sales” vs. “Wearing an ear-bud hurts the head.”
 Sentiment Analysis in action ..
• This is a case study regarding the performance of a bank in Johannesburg, South
Africa.
• Repusate.com, an umbrella company under the Sprout Social, specialising in
sentiment analysis and social listening helped the bank evaluate it’s customers’
sentiments.
• Repustate started a social media listening campaign. Over a period of 90 days, the
bank gathered around 2 million texts.
• Repustate discovered 7 categories of complaints that were relevant.
• Using named entity recognition the text was classified as belonging to one(or
more) of the categories.
• “I really hate going to my branch during lunch time and seeing only 2 tellers with a
huge lineup. Why do they go to lunch all at the same time?” Repustate's would
categorize this as belonging to the "Brach banking" category, and then applied
sentiment. The raw sentiment score for this text would be < 0, indicating negative
sentiment.
Sentiment analysis in action ..
 Sentiment Analysis in action
• Most negative tweets were related to branch, support and online categories.

• As an example, here were the most common phrases under the "Branch"
category:
o “Long lineups”
o “Always closed”(referring to hours of operation)
o “no tellers”

• After identifying the major problem, the bank worked in that area and a
same campaign was run 6 months later. 1 million responses came in via text
and social media. Repustate applied the same methodology and the results
were -
 Sentiment Analysis in action..

The magic of Sentiment Analysis.

Source : https://www.repustate.com/banking-sentiment-analysis-and-text-analytics/#banking-case-study