Simply put, for the purposes of social media, it is the process of determining the author’s opinion conveyed in a post.

The Importance of Sentiment Analysis in Social Media

Social media sentiment analysis can be an excellent source of information and can provide insights that can:

• Determine marketing strategy
• Improve campaign success
• Improve product messaging
• Improve customer service
• Test business KPIs
• Generate leads

In a nutshell, if done properly, social media sentiment analysis can improve your bottom line.  However, if you are making decisions using incorrect sentiment analysis data, the results can be catastrophic. Most social media analysis vendors will admit (if you push them hard enough) that their sentiment analysis algorithm will be, at best, 50-60% accurate.

[adinserter name=”AdSense 1″]

I certainly don’t want to be making business decisions based on 50% – 60% accuracy and I’m guessing you don’t either !  So what should you be looking for in a good sentiment analysis algorithm?

The study of sentiment analysis, if done properly, is exceptionally complex and is actually a field of study, not just a feature in a social media tool. I should probably be clear at this point, that the objective of this blog is not to discuss the nuances and detail of sentiment analysis. In fact, quiet the opposite, I am trying to simplify this very complex topic so that you can use the information when deciding on the social media tool or services that you need.

You probably recognise the complexity when you realise the number of times people misinterpret conversations or mis-comprehend the written word. Also, elements such as sarcasm and jargon make it even more difficult to determine meaning from words. So how do we simplify this topic so that we can evaluate vendor tools? There are a number of factors to take into consideration.

 [adinserter name=”AdSense 1″]

Types of Sentiment Analysis

Firstly we need to understand the methods that social media vendors use to determine sentiment. As I mention above, there are many types of sentiment analysis. However, for the purposes of this article we will concentrate on three:

1. Manual processing

Human interpretation of sentiment is definitely the most mature and accurate judge of sentiment. However, it still isn’t 100% accurate.  Very few vendors still use this process without the additional use of a tool. This is due to the prolific growth of social media.  According to Seth Grimes, social is the fastest growing source of enterprise analytical data. (Source: Seth Grimes Social Media and the Enterprise Business Intelligence/Analytics Connection, A 2010 Study conducted by Seth Grimes)

According to Yellow Pages: 69% of Australians use social media. (Source: Sensis).  Therefore, if you are going to use social media to determine sentiment, it is becoming less practical to use human processing and more likely you will need to automate the process.

2. Keyword processing

Keyword processing algorithms assign a degree of positivity or negativity to an individual word, then it gives and overall percentage score to the post. For example, positive words, great, like, love or negative words: terrible, dislike

The advantages of this method are that it is very fast, predictable and cheap to implement and run.

[adinserter name=”AdSense 1″]

However, there are numerous disadvantages including dealing with double negatives or positives, or different meanings of words, for example: the use of a word such as ‘sick’ (to mean either “ill” or to mean “awesome”). Not to mention, different researchers may assign difference percentages of positive or negative to word. More often the issue is that it does not deal with multiple word/context issues or non-adjective words.

Most vendors represented in Australia use a keyword processing algorithm.

3. Natural Language Processing

(NLP also called: text analytics, data mining, computational linguistics)

NLP refers to computer systems that process human language in terms of its meaning. NLP understands that several words make a phrase, several phrases make a sentence and, ultimately, sentences convey ideas. NLP works by analysing language for it’s meaning.  NLP systems are used for in a number of areas such as converting speech to text, language translation and grammar checks

(Source: mashable)

It can be likened to programming an algorithm to interpret the English language (or any language for the matter) with the rules that you were taught in English class.

Although NLP may seem to be far superior to keyword processing, it still has its limitations. Sarcasm a well known Australian trait, is very difficult to detect using NLP as is hyperbole (exaggerated statements) and social media acronyms (e.g. OMG, BFF, BTW etc) or social jargon such as:

• Youturn: To follow another person on social media with the intention of unfollowing them once they have you followed back, esp. on Twitter
• Wallflower: A person who regularly consumes the social media of others but never posts
• Face Crawling: Begging for Facebook likes, online or offline
• Hash-Browning: The excessive use of hashtags within a single post
• Metapals: Social media connections that have never personally met

People express opinions in complex ways for example: the difference between “I’m fine!!!” and “I’m fine.”. Also, changing topic mid post can be confusing. Below are some examples of difficult posts to analyse:

There are many uses of the word “sick”. The posts below show the use of sick in both a positive context and negative context:

Sometimes even human interpretation can be hard to determine. For example, does the Vine post below use the word sick to mean “cool” or “ill”?

…keep watching the video and you will see it is “ill”, not feeling well

[adinserter name=”AdSense 1″]

What does the future hold for sentiment analysis?

• We can assume that the future of sentiment analysis will plug the existing gaps in being able to interpret meaning.
• Increased accuracy when compared to human processing
• The ability to interpret human emotions: according to research by Glasgow University. there are six basic emotions of happiness, sadness, fear, anger, surprise and disgust.
• Improvements in machine learning accuracy source: B. Pang:(source: http://www.cs.cornell.edu/home/llee/papers/sentiment.pdf)
• Predictive analytics – once we have extracted sentiment and believe it to be accurate we can then predict future trends and behaviour.

Below are some interesting ideas from Phil Wolff via Quora: (source: Quora)

• Longitudinal analysis. Where are the cycles and patterns in sentiment? How does Phil’s attitude change during the day, year?
• Root cause analysis. What activities or people affect Phil and how? If we can see what Phil reads, where he goes, who he talks to, how much he moves, music he hears, what he eats, can we identify likely triggers for sentiment and affect changes?
• Realtime scoring. At scale, limited only by latency.
• Scoring reflect new models of cognition. Neuroscience and cogsci will inform what and how we measure, analys e and report. How likely is it that new social gestures will amplify patterns discovered through brain imaging?
• Analysis of non-textual inputs. Facial microexpressions captured in Skype chats, body language in YouTube videos, gestures in Google Glass, typing speed/interval/error patterns in Bing search, stress analysis in voice calls, clickstreams in Chrome, check-ins in FourSquare, physiology from Quantified Self  loggers – all will complement text analysis.
• Micropublic reporting. How are the people attending this meeting in two hours feeling now? This is aggregating sentiment for smaller, defined groups.
• Predictive sentiment analysis. How will they feel when the meeting starts in two hours? What are likely causes of drift?
• Sentiment streaming. Sentiment as real-time presence. Phil’s mobile emits a stream of Phil’s happiness, engagement, focus when he’s awake.

[adinserter name=”AdSense 1″]

Summary

Sentiment analysis is a difficult technology to get right. However, when you do, the benefits are great.

Look for a tool that has uses Natural Language Processing technology and ideally with machine learning capabilities. Look for a vendor that treats sentiment analysis seriously and shows advancements and updates in their sentiment analysis technology.

To receive more digital marketing and social media listening tips, information and templates sign up with Results 2Day

About Us

Results 2Day can assist you to find the right technology for your sentiment analysis requirements as well as provide reporting services focusing on sentiment analysis (generally at less than a monthly licence fee). To find out more, contact me at:

Christine.Day@results2day.com.au or (02) 8354 1554