Feelings, trends and value: Three key elements of sentiment analysis

Using OpenText Magellan Text Mining to spot and profit from human opinions

Do you want to know what your customers think about your product? How about validating why and when your employees started to complain about the new intranet logging policy? Or perhaps you’d like to capture how the mood in social media is changing before an upcoming referendum? With web and social media crawlers – tools to connect to content and transform content to an input – sentiment analysis can help you to easily undertake large-scale opinion analysis on almost any content online.

Part of the OpenText™ Magellan™ AI-enhanced analytics platform, the sentiment analysis module of Magellan Text Mining identifies which statements are subjective and which ones are objective. The Magellan Text Mining sentiment analyzer can highlight the opinions around various entities, such as people, names, products, organizations and so on. Furthermore, it understands the major languages English, German, French, Spanish and Portuguese.

But what does it mean to analyze the sentiment and what’s the value of this? Here are three things to remember when analyzing sentiment using Magellan’s native text mining module.

1. It’s about subjectivity

The content you’re performing sentiment analysis on must be subjective in nature to make the results useful. Many types of content, such as financial reports, technical specifications, scientific articles, tax forms or user manuals don’t lend themselves well to this process because the authors of these documents are trying to be as factual and objective as possible.

Other content types, such as blogs, reviews, comments, tweets, e-mails, customer care tickets, newspaper articles or even resumes will include statements of opinion or emotion – whose position, tone, and intensity could provide useful insights. Hence, we want to focus on contributors such as product reviewers, blog authors, social media users.

Why do we want the software to point out opinion and emotion, things that by definition exist only in someone’s head and don’t necessarily have a rational basis? Well, subjective opinion drives many personal and even organizational choices, and emotion is a useful indicator of what people care most about. Let me give you an example: If I look at customer care tickets, there are usually some issues at the beginning that a person would like to resolve. So we might assume this person is simply complaining; it could be about a software feature or a hotel’s accommodation, but it always starts with an opinion. Our software can spotlight the kind of opinion, i.e. the tonality – positive or negative.

The sentiment analysis feature within Magellan Text Mining goes beyond basic text processing that simply compiles what customers are saying about a feature and maybe highlights the most common terms. It guides us to the aspects of the software feature or hotel stay that matter most to customers and how they really feel about it.

2. It’s about trends

Our sentiment analysis tool will expose subjectivity and tonality for each sentence within any piece of text. When the text contains opinions, there will be positive, negative or neutral ones. There could be also both positive-negative statements when the module captures ambiguity. Applied on a large set of reviews, it will show whether the overall trend is positive or negative.

This is natural language processing, which can handle irony and double meaning in text. There may be allusions to contexts we cannot know if we are not sitting next to the person writing. For example, it might turn out that someone who’s saying “This is really a great product” was being ironic. Or it might happen that a “fair” price in a review was fair for the reviewer, but would give someone else sticker shock. Therefore, the analyzer should be seen as a tool to show overall trends and moods, not to accurately capture individual opinions.

And yes, it will compile both subjectivity and tonality for the entire piece of text, but it might be wrong when it comes to something very granular, such as when a negative statement is removed from the context and appears positive instead.

3. It’s about value

With our sentiment analysis tool, the real value lies in statistics. If you read only two reviews of the hotel prior to your next vacation, can you really feel confident that they judged it accurately? But when you can see a score for your hotel based on 5,000 reviews, most of which rated it as a nice place to stay, you can feel confident in the quality of the hotel.

The same thing happens when you look at our tool’s results: The overall emotional value derived from thousands of statements written by thousands of users will give you the real picture. In the next step, these results can be combined with topics, persons, products, keywords, revenue indicators, dates (and so on) to show the true insights.

Using Magellan dashboards, we can use this module to provide real numbers and even apply algorithms to correlate positive or negative scores with specific information coming from a database.

For example, we can expose why and when communications between a support team and a customer turned from positive to negative or we can identify the tonality around different topics pertaining to your products (see image below). We can show which topic was well accepted by the audience and which one the audience had no interest in. Now, that’s the real value!

In future blog posts, we’ll present more about how to realize value from sentiment analysis and other forms of content analytics through our Magellan platform.

Learn more about OpenText™ Magellan™ or OpenText’s AI & Analytics Service Practice, or email us for a demo.

Are you ready for the Intelligent and Connected Enterprise? Join us at OpenText™ Enterprise World 2019 to hear how we’re enabling the Intelligent and Connected Enterprise with AI and the Internet of Things.

Robert Kapitan

Robert Kapitan is a Senior Technical Consultant at OpenText working for Professional Services. Robert has been working with Text Mining applications for last 20 years helping to build software solutions that will understand human language. He holds an M.A. in Theoretical Linguistics and a Ph.D. in Cognitive Semantics.

Related Articles

Close