Read on, little robot.

How can machine learning boost media monitoring?

In one of our previous articles titled “Giving media analysts some free space”, we described the goal in media monitoring as to gather relevant information to a specific topic, product, company or organization from various heterogeneous channels. In order to improve the quality of the filtered information, Media Response Analysis is brought into play.

In the same article, the benefits of automatic text summarisation techniques to professional media monitoring service providers and media analysts are presented. Automatic text summarisation is considered a very important area within machine learning. In general, during the last years there has been an increased interest towards automating the media monitoring process through the proper utilisation of machine learning approaches. But what exactly is machine learning? And what kind of solutions can it provide to the business problem of media monitoring?

Machine Learning
Machine learning is a field of computer science, which involves the research and development of algorithms that have the ability to learn from and make predictions on data. A definition provided by Arthur Samuel back in 1959 sums up the underlying theme of machine learning: “Machine learning is the field of study that gives computers the ability to learn without being explicitly programmed”. Of course, since 1959 many things have changed in this scientific discipline. Especially nowadays, with the constant emergence of new computer technologies, machine learning, already a very extensive field, is continuously evolving and expanding, providing applications and solutions to a broad range of real world problems.

Machine learning: How to get smarter machines?

Machine learning: How to get smarter machines?

Machine learning: How to get smarter machines?[/caption] Although machine learning tasks can be distinguished into many types, two of the most widely adopted categories are “supervised learning” and “unsupervised learning”. In the former, the algorithms are first trained using a set of labelled examples, i.e. inputs, for which the desired outputs are known. The goal for the learning algorithms is to acquire the ability to successfully generalise from the training data to new, unlabelled inputs. In the latter, on the other hand, only unlabelled examples are available, leaving the learning algorithm on its own to discover hidden structure and patterns in the data. Next, we will talk about classification and cluster analysis, the most well-known representatives of supervised and unsupervised learning, respectively and their application to media monitoring.

Classification

Like or Dislike? Find out with Sentiment Analysis

Like or Dislike? Find out with Sentiment Analysis

Classification investigates the problem of assigning observations to a predefined set of categories (also known as classes). As already mentioned, classification is considered an instance of supervised learning. Therefore, the algorithm that implements classification (the classifier) is provided with a set of correctly classified instances and trained, in order to be able to correctly predict the category membership of new, unknown observations.

Classification can prove really useful in Media Monitoring and Media Response Analysis applications. For instance, let’s suppose that a client needs the information gathered by the media monitors classified into a desired number of predefined categories (e.g. Economy-Business-Finance, Lifestyle-Leisure, Politics, Science-Technology, etc.), for better organisation and easier visualisation of the results. Another interesting scenario is the case where there is a need to evaluate opinions and statements found within news articles, social media posts, etc. (this task is also known as Sentiment Analysis). The abovementioned tasks can be easily and efficiently addressed with the use of properly trained classification models.

Cluster Analysis
Cluster analysis, also known as clustering, deals with the task of identifying groups (called clusters) of objects that are similar to each other but different from objects in other clusters. It must be noted that cluster analysis does not refer to a specific method or model, as many different algorithms and approaches, encompassing different notions and concepts can be employed, depending on the needs at hand.

In Media Monitoring, a very interesting task is the discovery of topics or events in large volumes of information collected by the media monitors for their customers. This can be tackled as a clustering problem and the whole process can prove quite helpful, as it allows for an extra filtering and grouping of the gathered information and reveals patterns (topics / events) within the data that are not evident in the first place. However, several challenges can usually be implicated in this task, as the number of topics / events may not be known a priori and the data can be highly noisy. Hopefully, with the combined use of algorithms that are able to extract the correct number of topics (clusters) and feature selection methods, it is possible to overcome these challenges quite easily.

From the abovementioned, it is becoming obvious that the use of machine learning approaches (not limited to classification and clustering techniques) offers tools and solutions that can facilitate the daily work of media monitors and analysts, allowing them to increase the amount of automation in the workflow of their companies, with high-quality services being provided to the clients as a result.


Photo Credit: