People have always had an interest in what other people think, or what opinion they hold. Since the inception of the internet, increasing numbers of people are using websites and social media platforms for expressing their opinion. Due to platforms such as Facebook, Twitter, etc., it has become feasible to analyze and extract public opinion on a certain topic, news story, product, or brand. Opinions that are mined from such services can be valuable. Data mined from these sources can be analyzed and presented accordingly to easily identify the online mood (positive, negative, or neutral). This allows individuals or businesses to be proactive as opposed to reactive when a negative conversational thread is emerging. Alternatively, positive sentiments can be leveraged to identify product advocates as well to shape the business strategy by seeing the parts of the strategy that are working.

Sentiment Analysis Techniques

There are many ways in which Sentiment Analysis can be implemented but since essentially being a text classification problem, it can be broadly classified into two main areas:

1. Supervised Learning

In this technique, a classifier is constructed and the problem at hand is studied intensively. The role of the Classifier is to analyze texts and categorize them into a positive, negative or neutral polarity. The three main classification techniques used in this are Nave Bayes, Maximum Entropy and Support Vector Machines(SVM).

2. Unsupervised Learning

This technique is more of a sequential tagging algorithm which follows a set of steps to identify the sentiment behind a piece of text. It first implements POS tagging, then, two consecutive words are extracted, their tags are identified to check if their tags conform to given patterns. Then, the sentiment orientation of the extracted phrase is estimated and then the average of all the estimated SO of all phrases is computed to giver the overall sentiment orientation of the piece of text.

Challenges with Contextual understanding & tone

As fancy and advanced it may seem, Sentiment Analysis is not a perfect science at all. One-dimensional prediction of sentiments is fairly easy but when things like contextual understanding come into play, then we still have a long way to go. Due to the complexity of human language, teaching a machine to analyze it with different grammatical errors, slangs, misspellings is a difficult process. It makes it even more challenging when cultural differences come into play. Teaching a machine to understand how tone is affected by context is even more difficult. Also, when humans interpret a tone, it is also dependent on the person’s intuition and that cannot be replicated. For example, consider the following sentence, I broke my phone. Wow!. Now, most humans can easily interpret that the person is being sarcastic because we know having a broken phone is not a good experience for most people. By applying contextual understanding to this sentence, we can easily infer that this sentence has a negative sentiment.

Sentiment Analysis Applications

The applications of Sentiment analysis are broad and powerful. It is being used by many companies to enhance their customer experience as well as products & services. Some of the popular applications by companies are:

• Expedia Canada took advantage of sentiment analysis to quickly understand consumer attitudes and react accordingly when they started receiving multiple negative reviews for the music they had used in one of the advertisements. People were venting out their frustration on social media about this which was leveraged by Expedia to know the sentiment of the response and act upon it.
• Almost all companies have their own Facebook pages & twitter handlers and get tweets, comments, reviews and complaints on the platform from their customers. They employ sentiment analysis to track those comments and tweets, analyze them to extract the general sentiment and work upon it to retain their customers.
• Some companies also employ sentiment analysis algorithms in their corporate network. For example, a sentiment analysis algorithm could be used to filter the incoming mails based on the tone of the mail. Outlook also provides a Tone Detector Add-in that determines the tone of the email as you type.

What the future holds?

Still being relatively immature at the moment, it is really tricky to speculate what the future will hold for sentiment analysis. One general assumption says that sentiment analysis must move beyond a dimensional sentiment scale i.e. positive and negative. For the future, to truly understand and capture the broad range of emotions that humans express as the written word, we need a more sophisticated multidimensional scale which takes into account all the various emotions a human is capable of expressing through words. Until we can measure skepticism, hope, anxiety, excitement, sentiment analysis is (literally) and will remain to be one-dimensional!

How can you use Sentiment Analysis data in Salesforce?

Salesforce NLP (Natural Language Parsing) services can be used for Sentiment Analysis. You can create a training data set and create a model based that can tag different interactions through social channels, phone text, or emails with sentiments. Call us at 855-MIRKETA if you need help in getting started with your Sentiment Analysis project.