Sentiment Analysis

Data analysis

What if you could find out what your consumer base thinks about the launch of your new product without having to run multiple surveys and customer evaluations? How about if you could see how your event/festival attendees are feeling during the event; allowing you to make adjusts accordingly, not having to wait until next year? These are concerns, whose clarity would greatly improve the continued support of a product or service we are providing without having to wait for consumer reports and final sales numbers. There is a solution to these concerns that can help improve product launches, event management, and overall consumer perception of your company and its brands.

One of the most coveted resources to arise in the last decade is data. Marketing researchers across all industries have been drawn to this vast wealth of information openly available. As a whole, data researching has established itself as a lucrative endeavor with a wide array of different subsets lending themselves to the overall interpretation of raw data. One of these subsets is sentiment analysis.

What is Sentiment Analysis?

Sentiment analysis is a data researching technique that utilizes building systems that can identify and extract opinions from text-based data. Essentially, an algorithm is created that searches for specific phrases and keywords in a pool of text-based data. By using sentiment analysis, you are able to construct a system that can be updated and manipulated in order to improve accuracy and apply to a variety of different objectives.

Sentiment analysis works because of the data it utilizes. Text-based data is vital to a sentiment analysis model as it is where all of the information is held and easily presented. Text is uniform, it’s presented in a standardized way where the only variables are the words held within them. There aren’t any photos, videos, or other media types that come in all sorts of files sizes and types, making it easy to program an algorithm to interpret it.

Another positive of using text-based data is that it can be broken down into fact and opinion. In order to find out about your consumer perception it will be important to focus on those opinion-based data points. While opinions can have a wide reach due to their personal viewpoint, it’s important to remember that sentiment analysis looks at the overall point of view, drawing a conclusion from the majority attitude. Giving opinion-based data an honest look can be a harsh reality for some companies but it’s important to keep the objective in mind; getting honest, in the moment feedback from consumers.

Text-Based Data and Why We Use It

You might be thinking “where do I get this text-based data from?”. Well, you don’t have to put too much effort into collecting data or paying some company to provide it to you. The internet openly provides plenty of data that’s just waiting to be analyzed. Most sentiment analysis data is sourced from public platforms such as reviews, forums, and social media. Twitter is by far the main supplier of sentiment analysis data as it is all text based and publicly available.

These different types of venues are great for gathering data because they are supported by the public as well as easily available. On these kinds of sites people are sharing their honest, in the moment, opinions.  Unlike a research database where the data may be skewed by how the surveys were questioned and who the audience was, public text data is supplied by those who aren’t even aware that their contributions could have an influence. With sites like Twitter or Yelp, the only barrier to entry is an internet connection and a will to share your opinions. This allows for many different types of voices to be represented without you having to actively seek them out.

How Sentiment Analysis Works

When it comes to actually creating a sentiment analysis module there are a few things to consider. We won’t go into full detail about our methods but are happy to give you a better idea of how raw data turns into results.

There are a few major steps when it comes to sentiment analysis, first is to know what you want to find out. If we were to take the example from earlier of gaging attendee perception of the event you are running, we would want to focus on finding testimonials that show a positive reception. Most likely you’ll be looking for positive experiences but it’s important to keep in mind any specific aspects of that positive experience that you might be looking for.

Sentiment analysis can be done in two different ways: rule based or automated. In rule-based sentiment analysis specific actions have to be laid out in order to render the results you wish. A specific lexicon is established and these keywords are matched with any correlating words in the data. An example of this will be walked through below in order to get an idea of how sentiment analysis, as a whole, works. Automated sentiment analysis is a machine learning pattern algorithm that can predict words from the sentiment. This process improves on itself, learning how each element of the data should be valued. Automated sentiment analysis can take some time to perfect but is often more accurate. That being said rule-based sentiment analysis, while often less accurate, is easier to adjust and update if you are looking to change objectives of your evaluation.

The Steps of Rule-Based Sentiment Analysis

Most of the time, sentiment analysis is done through Python, as it is a well known programming language that can easily accomplish what we are looking to achieve. TextBlob is a Python library that processes textual data and what we will be using in this sentiment analysis example. In order to walk through the sentiment analysis process, we will be using the example text of “I really enjoyed this conference!”.

The first step of the sentiment analysis process is tokenization. This is where you break down a data point into its individual elements. In this case we would make each word its own element, as well as the exclamation point.

Second, we want to clean up the data. In this step we want to remove any excess information that might throw off the algorithm. In this case this would be the exclamation point. At this point it’s important to remove any symbols, punctuations, emojis, or links that might be included in the data.

Third, we want to remove any stop words in the data. Stop words are anything that isn’t relevant to our overall objective of assessing consumer opinion. In this situation we will remove “I”, “really”, and “this”. Depending on your specific thesis you might consider keeping “I” as it directs the data towards a specific individual, but we will remove it for this example. “Really” is removed as it can apply to both negative and positive experiences. Something can be “really good” as well as “really bad”. If we were to include “really in the analysis it could throw off the validity of the results.

Then we will classify the remaining elements as either positive, negative, or neutral. In this case we would categorize “conference” as neutral and “enjoyed” as positive. If something is positive, we will give it +1, if it is negative, we will give it -1, and if something is neutral it will receive a zero. In this situation we would have a +1 and a zero.

Afterwards we calculate the total of the data point. In this situation we have a total of 1, a positive score. Obviously, this is a simplistic example but data such as reviews will usually list their pros and cons allowing for a more volatile analysis.

Sentiment Analysis as a Decision-Making Tool

Now that you have a better idea of how sentiment analysis works you might be thinking of how it can be implemented into your business. Aside from the major examples of a product launch or event management it can also be used as a benchmark of how your company or brand is doing on a regular basis. You might not always have overwhelming feedback about your every decision like some large corporations who find themselves in the headlines due to an assortment of issues such as sales numbers, controversies, or announcements. Sometimes you need a checkpoint analysis of where you stand in the market. If you are working on a competitive intelligence analysis, sentiment analysis can be a great way to get a true consumer perception testament for your company as well as your competitors and industry as a whole.

When it comes to data research, sentiment analysis can offer vital information into how you are being viewed by the public at large. With so much of our time being spent online and on social media our opinions shared on these channels should be held with a weight that is substantial compared to that of surveys and impressions given offline.