Protect your business, your brand and your reputation more by adding sentiment to your media monitoring.
We often discuss the large volumes of media content that exist today and the difficulties in unearthing quick, actionable value out of such an overwhelming mass – but sometimes it’s the simplest things that add the greatest value. Take sentiment analysis for example – recently introduced to our Mediaportal.
For those who are unfamiliar, sentiment analysis is the process of computationally identifying the tone within a piece of text in order to determine or categorise whether the attitude expressed is positive, negative or neutral. It is by no means definitive, but can be incredibly valuable in identifying potential issues or successes when viewed at an aggregate level, or simply when viewed within a topic or event.
“Your brand is what other people say about you when you’ve left the room”
Jeff Bezos, Founder Amazon.com
While the applications of sentiment analysis are broad and powerful, it’s often spoken about more in terms of brand – and specifically for Marketers. However, as Communications roles intersect, and media types diversify, it’s incredibly timely to explore the potential that sentiment holds to improve crisis management tactics, social media management and more through what is really ‘opinion mining’. Following our sentiment release, we spoke to our team around how we approached introducing this feature across our platform – and some of the complexities that are at play.
How we went about introducing sentiment to Mediaportal
Sentiment in Mediaportal is derived from both Google Natural Language Processing and an open-source toolkit made available by Stanford University. This has several advantages over a simpler approach, most notably that they are capable of taking the structure of a sentence into account when forming a judgement about sentiment for the purposes of categorisation.
For example, a simple approach may judge that the phrase “not bad” is negative because it contains two negative words. While in reality, one word (“not”) is simply telling the reader to think the opposite of the next word (“bad”). This and other similar examples are accounted for when using a technique that takes advantage of not just word choice, but also sentence structure.
To put it simply, what you say and how you say it are equally important. Using existing and well-developed services allows us to provide our customers state-of-art performance and the widest coverage possible from the very start of our sentiment releases.
The shades of grey of sentiment
The difficulty with sentiment is when it becomes hard for automation to identify or understand the nuances that exist within language, such as sarcasm or local jargon – and of course how to apply ‘context’.
It’s very possible for individuals to have differing opinions of the sentiment of a media item. For example, an item may contain strongly negative content but contain a positive mention of an entity of interest – such as your organisation. In this case, you may view this to be in fact ‘positive’, whereas your competitor may see this as ‘negative’ as it reflects overall sentiment against the industry that they’re also a part of.
Another recent example can also be seen in the ABC’s article on a ceremony to celebrate the Thai rescue cave divers.
- Positive sentiment “The Australian doctors and divers involved in the Thai cave rescue will be formally honoured today at a special ceremony at Government House in Canberra.”
- Negative sentiment “Just hours after the rescue mission was completed, news came through that Dr Harris’ father had died while he was inside the cave.”
While the overall article may be made up of primarily negative language, there may be pockets that could be viewed as positive which is why we have focused on generating sentiment scores for a media item as a whole – taking more than just the headline into account.
While sentiment can be easier to determine on things like movie reviews given the nature of what is asked –good or bad – alongside additional information like a star rating, the complexity in sentiment analysis extends beyond media content. To provide greater accuracy, the first release of sentiment includes the ability to override the sentiment badge on each item so that our clients have full control over their coverage.
Analysis is a big part of the next phase for sentiment as it can be incredibly helpful when reporting on efforts like ‘firefighting’ a social media crisis, or simply extracting the tone of your overall customer experience to determine next steps.
It’s also not limited to media items, sentiment has the potential to help identify who may be driving a particular tone and uncover trends of a skewed perspective.
To learn more about how we’ve integrated sentiment into Mediaportal, click here.
Marketing Specialist – Lead Gen at Isentia
Louise is an experienced content marketing professional who translates Isentia’s marketing strategy into impactful and effective marketing campaigns across multiple channels. As the Lead Gen Marketing Specialist for Isentia, Louise enjoys creating informative and engaging content for media and communications professionals.