After our successful kickoff meetup in Leuven, we invite all interested NLPers to get together in Antwerp on February 21st 2017. This time we'll hear from three companies whose core technology relies on Natural Language Processing. Miia develops technology that understands texts in complex fields such as recruitment and legislation, Utrecht-based Blendle has taken the media world by storm with its "iTunes for news", and new kid on the block Textgain offers web services for predictive text analytics.
• Predictive Text Analytics & Profiling
Guy De Pauw, Textgain
Interest for Big Data applications has never been higher. Many companies are unlocking the value of their data and gaining useful insights by analyzing large, structured databases of customer records, financial transactions and the like. Yet, most of the interesting digital content is locked in an unstructured format: language. Never before has so much content been written, more than we could ever hope to read in a lifetime. But the knowledge that is contained in this continuous stream of unstructured data seems beyond our grasp. Unless through the use of text analytics! In this presentation we will show how you can quantify facts and opinions on a global scale through machine reading. What's more, this technology can even "read between the lines" and reveal data about your target demographic through author profiling techniques. We will outline some of our prior projects that may inspire you to integrate this technology in your organization as well.
• Natural Language Processing at Miia
Stephen Lernout, Miia
Miia developed a proprietary AI system enabling machines to understand text and natural language. Finding meaning is done by analyzing text linguistically, mapping words and expressions onto a ConceptNet and using powerful semantic pattern recognition to combine these concepts into meaningful entities. Machine Learning is applied to build these ConceptNets (semantic brains) automatically and APIs are in place to integrate and scale the Company‘s Natural Language Understanding products rapidly. The Miia engine works language agnostic and is domain independent.
• NLP at Blendle for Personalised Recommendations
Martijn Spitters, Blendle
At Blendle, every morning we receive over 6000 fresh articles from the latest newspapers and magazines. Our main task at that point is to compile and send out personalised article bundles to hundreds of thousands of users. Because at that moment the new articles have not been read by any of our users, we are dealing with a huge cold-start problem and can therefore not solely rely on collaborative filtering-based recommendation techniques, nor can we use the popularity of articles as clues for what our users may want to read. We tackle this cold-start problem by a mix of curation by our editorial team and an automated analysis of the article content. We apply an article enrichment pipeline, which contains a wide range of NLP modules for a.o. named entity extraction, topic modelling, stylometric analysis, semantic linking, and even predicting the pick potential to support our editorial teams in their article selection process. In this talk I will go into detail about our enrichment pipeline and the algorithm we use to create our personalised bundles. I will also show how our batch processing and streaming architecture is set up.