Machine learning and what it means for media intelligence
By Michael Bade
For years, the media intelligence industry has provided communicators with tools to identify and statistically analyse, to some degree, what is being reported about themselves and their industries in social channels, alongside their traditional media.
However, with SO much social data available (as in, tens of millions of social posts published across Twitter, Facebook and Instagram every minute) there has been a limited ability for communicators to really “do anything” with the conversations being captured in social media in the same way that we have with traditional media.
It’s the evolution of artificial intelligence (AI) and specifically, machine learning, that has enabled us to really make sense of the huge volumes of data by capturing and analysing that data, as it is published. This is critical for social media due to the volumes of data, but has also had positive implications for traditional media in terms of gleaning deeper insights in a timeframe that is meaningful and actionable.
So, what is machine learning and how is it differentiated from AI?
I touched on this in an earlier blog post, and the best definition, as stated in the Oxford Dictionary, describes Artificial Intelligence (AI) as the theory and development of computer systems able to perform tasks that normally require human intelligence. Machine learning is a sub-field of AI whereby computers learn from data without the need for strict directions. The first step is to create tasks or examples to machines and provide data for the tasks or examples to be applied to. The machines identify patterns that emerge in the data, and improve their processes automatically through data experience and with minimal human input.
Most people experience the results of machine learning on a regular basis. Fraud detection on your credit card that doesn’t match your regular spending habits, online shopping recommendations for consumers, and Facebook prioritising content on your newsfeed are all examples of machine learning at work.
Building on that general understanding, there are a number of different types of machine learning to consider (stay with me)
- Supervised machine learning is where the task of learning is directed by input/ output expectations. The machine is given labelled data and it seeks to identify instances of those data. For example, seeking pictures of elephants as labelled by humans, and the machine learns a way to identify pictures of elephants on its own. Supervised learning is typically assessed for quality by retaining some of the trained or labelled data for evaluation purposes. Quality can then be assessed by running the newly created algorithm against the evaluation data.
- Unsupervised machine learning is whereby no labelled data is input and machines seek to find patterns in data that are previously unknown. For example, identifying specific types of consumers’ buying behaviour based on a large set of sales data.
- Semi-supervised machine learning combines a small amount of labelled data with a large amount of unlabelled data and can result in significantly more accurate learnings.
- Reinforcement machine learning trains machines by giving feedback on its data categorisations and decisions to provide improvements on future learning.
In media intelligence, most machine learning typically uses supervised learning and is performed against unstructured text (the contents of news and social media posts). This means that people train the engine to perform a particular result such as:
- Categorising text to industry verticals
- Performing sentiment analysis
- Junk detection
- Recommended content detection
- Entity detection
This “training” is often performed internally but is also performed by clients as they use the platform in their everyday tasks, for example, as they perform sentiment overrides or determine “junk” items. Even the simple process of allowing a machine learning algorithm to assess the media that you’re engaging in every day helps to train the engine on the types of content that are relevant to you.
So how does machine learning make media intelligence “better”?
Determining insights from social media is an ideal application of machine learning as it relies on huge volumes of data to hone its accuracy, with the many conversations happening online which are otherwise complex to digest and derive meaning from.
Machine learning enables communicators, in near real-time, to not only understand what is happening in their media coverage, but also to cut through the “noise” and understand emerging crises, trends, or to predict future behaviour of consumers and journalists, empowering communicators to be on the front foot. Some of the ways machine learning achieves this is via:
- Identifying entities within articles – people, businesses, events, data, brands - and applying logic that provides a robust sentiment analysis of those entities, as well as the sentiment of the article in its entirety.
- Identifying influencers relevant to a client – those who are most likely to reach, engage and motivate your audiences and stakeholders – can be identified by analysing topics they write about.
- Alerting clients to recommended content based on their viewing and engagement behaviour and equally, identifying “junk” items based on learning from previously junked items.
- A better understanding of increasingly visual social media, such as Instagram and Pinterest, recognising logos, people and items in imagery.
The application of machine learning to media intelligence results in more informed communications, faster. It gives communicators the ability to use their data in more flexible ways alongside their traditional media, for deeper analysis and reporting to stakeholders, and it means communicators can be more proactive rather than reactive. Machine learning is smart humans and smart technology synergising to drive communications technology forward in leaps and bounds.
If you’d like to find out more about how machine learning can benefit your organisation, contact us at hello@truescope.com
