Artificial intelligence (AI) and machine learning are here and reshaping business. But AI isn’t just for tech companies and online retailers—it can be a force multiplier for smaller and mid-sized information companies by enabling them to take control of content and audience data by automating many of the manual tasks associated with maintaining that data and helping create personalized and meaningful experiences for customers.
At the upcoming Business Information & Media Summit, Vladimir Petkov, chief technology officer at A Data Pro, will offer an overview of the potential of AI and machine learning for information companies, as well as what information companies need to do to develop an AI-enhanced workflow.
Connectiv: Vlad, what are some of the practical applications for AI and machine learning specifically for information companies? What are some of the tasks that AI can perform that free up the human element?
Vladimir Petkov: Considering the endless ocean of data and information in the Digital Age, it is hard for media and information companies to stay on top without sinking into the deep mud of content. In our day and age, social media more or less controls the influx of information to news providers, stating what is important, setting trends and defining what readers care about. Artificial Intelligence (AI) is set to completely alter the landscape of modern media, the way we perceive it and the way information behaves in general. It has the potential to change the way journalists approach content, to make editors more effective and relevant, and to allow readers and consumers to choose and manage the content that best suits their interests or information needs. If media companies want to affect how content is created, developed, presented and delivered to the public, they need to take a proactive approach to implementing AI in their day-to-day tasks. In order to stay relevant, media companies need to start embracing the possibilities of machine learning (ML).
A Data Pro’s ML and human-in-the-loop (HITL) approach has many practical applications for media companies and information providers, whether of business data or otherwise. One of them is to optimize internal work processes through automation of manual, time-consuming and tedious tasks, by integrating different types of data into one inseparable whole, where all the information is accessible with a single click. Our solutions can automatically tag and classify articles, analyze their sentiment, extract key information and then connect the dots between them. This gives the company a better understanding of its own internal processes and streamlines the day-to-day work process.
Furthermore, in order to provide tailored, engaging content in the Digital Age, media companies must understand the way data is constantly being changed and manipulated. Our AI-powered solutions can provide round-the-clock updates on what is trending, thus freeing journalists from the task of doing manual research or media monitoring.
Sentiment analysis and countering disinformation is also a strong side of automation and AI. They can analyze the tone of voice of any given article, and identify manipulative content. Media companies can thus filter out irrelevant articles easily, while refining their own and staying on top of the never ending stream of data.
Connectiv: Your presentation at BIMS will include a case study on “human-in-the-loop” machine learning for SeeNews, a major European business news company. Please offer a brief overview of what you mean by “human-in-the-loop machine learning” and how SeeNews was able to leverage it?
Petkov: The HITL approach to machine learning is an extremely powerful way to approach big data management. It is unique in the sense that it intertwines the expertise and decision-making process of humans with the efficiency and speed of AI and automation solutions. The AI processes the data, automatically assessing it and making the connections, but machines are still not intelligent enough and make errors. This is where the human enters the loop - the process of automated data management is constantly supervised for errors by an analyst or data scientist, who can then carry out a quality check of sorts that double-checks the machine’s decisions, fine-tunes the end results and improves the process. But that is not all - the machine learns from the human’s input, constantly getting better at its job. This is a continuous process, and this is why the HITL model is key - the AI learns and improves every single time.
We put this approach to work when SeeNews, a business intelligence provider for Southeast Europe, approached us with the task of aligning its archive of more than 400,000 articles with its database of 4,000 company profiles. We then had to link them to the news stories they created on a daily basis, so that all new content was immediately linked to all the information already in their archive. Doing this gave readers the ability to immediately access further information about the companies mentioned in the articles, such as ownership, sales, executives and locations.
We did this through a Named-Entity-Recognition technology, combining it with the HITL approach in order to process the news archive and automatically identify names and profiles, thus freeing journalists from having to read the articles. The machine learning process was constantly overviewed and moderated by human input, thus countering any errors that the machine model might make.
Also, thanks to their deep content understanding, journalists could now work in real-time within a tailored, AI-powered content system, and train it over time to achieve even better results.
Connectiv: What does an information company need to be prepared to do to successfully integrate AI and machine learning into their workflow? Is the opportunity only for larger, more technologically advanced organizations or are there opportunities for smaller organizations?
Petkov: The beauty of ML and the HITL mechanic is their agility. Every project, for every client, can be approached as per its specifics, tailoring the approach without exception. This is true for media and information providers, both big and small, as well as companies providing business intelligence. To be prepared for this change, companies must be ready to provide initial information, or what we would call dataset, which can then be “fed” to our models so as to prepare a project-specific, bespoke solution to the data. The “one model fits all” approach does not work.
Providing an initial set of data is a very powerful way to initiate the process, as it gives us a solid foundation to work with. Other than that, companies need to invest in training and prepare their journalists and other personnel dealing with data, so they can continue working with the model and further increase its accuracy. This is the right place to say that these AI models do not threaten the involvement of human intelligence in the work process. On the contrary, they reassure it.
To sum up – companies must prepare their employees and provide the initial trainings needed for working with the AI and making it better with time.
Connectiv: If the initial opportunity today is the offloading of manual tasks, what will the next stage of AI and machine learning look like for media and information companies?
Petkov: Apart from the offloading of manual tasks, involving AI in the work process of media and information companies can provide, in the near future, fully automated text generation and content creation, which will further rid journalists of writing, for example, updates on financial data and business metrics. Our AI already automatically creates content related to mergers and acquisitions, and writes articles on topics like finance and company data.
In the coming years, AI will most probably also have the ability to foresee economic changes or market movements (to a certain degree, of course), and identify trends and topics ahead of their time, i.e. before they have even formed.
The approach can also be applied to tackle the fake news phenomenon, which is becoming an increasingly alarming trend, spotting disinformation and manipulative content. A well-trained HITL model can even detect and analyze hidden biases in reporting and storytelling, for example, if certain people, topics or industries are gaining more exposure than others.
What kind of content worked before? What content works now? What are the best practices? Machine learning has the potential to give answers to all of these questions, increasing or widening the audience of the company, and allowing it to better understand its own internal processes.