Chat Classification and Controversy Detection


This Friday we'll have two talks followed by drinks. Our industrial speaker is Gianluigi Bardelloni, Data Engineer/Scientist at the KPN NL D&A Data Science Lab. He will talk about customer service call/chat classification at KPN. Our academic speaker is Bob van de Velde, is a PostDoc who does research into personalization, political communication and controversy using social and computational science. He works with diverse teams from Communication Science (ASCoR), Information law (IViR) and information retrieval (ILPS) and is lead developer on Robin, a personalized communication data collection platform. He will talk about controversy detection.


16:00-16:30 Gianluigi Bardelloni

16:30-17:00 Bob van de Velde

17:00-18:00 Drinks and snacks


Gianluigi Bardelloni - Customer service call/chat classification at KPN

Correctly classifying transcriptions of customer service calls/chat is crucial to identify bottlenecks in our internal business processes and improve our customers satisfaction. The task is quite challenging due to noisy transcriptions and poorly labeled data. We will show how we are trying to achieve satisfactory topic modeling using both supervised and unsupervised learning as well as by defining a specific domain ontology. Keywords: CNN, RNN, LSTM, Doc2Vec, Word2Vec, k-means, ontology, topic modeling

Bob van de Velde - Controversy detection: Language, interaction, other?

Controversy begets conflict, requiring moderation. Yet the detection of controversy is not clear-cut. In this talk, I'd like to provide a brief overview to approaches to controversy detection with particular respect to non-Wikipedia pages. State of the art methods now employ mainly lexical approaches or leverage the platform-specific features of Wikipedia to classify controversy in outside domains. Can detect controversy using linguistic cues through lexical or semantic approaches, or do we require some form of interaction data?