Marie Skłodowska-Curie Actions

H2020 Marie Skłodowska-Curie - Individual Fellowships (IF)

    31/05/2018 17:00 - Europe/Brussels
    H2020 / Marie Skłodowska-Curie Actions
    Portugal, Lisbon
    Talkdesk Inc

Our goal in this project is a multi-lingual event extraction, event understanding and event relationship discovery from vast number of messages coming to a contact center and in the conversation between the customer and agent. These messages could be in any format including but not limited to speech, text, video and email. Such capability needs to run on existing 100+TB of data to extract events and to be able to run the same capability in real-time.

Event understanding in human being conversation as a new feature, needs to operate on existing TB of data and events data hidden in voices to identify and discover relationships between those events with the goal of identifying event patterns over time in customer communication and across, contact centers and geographic locations.

For instance, if a customer contacts a call center and says “when I was playing with the black button on top left suddenly the screen went all white and the off and on button does not work anymore on my tablet but it light keeps blinking”. As a result of this challenge we would like to extract the following insights:

  1. Pushing back button on top left of a phone > screen goes white
  2. Identify the phone with black button on top left
  3. Screen goes white > off and on button does not work + light keep blinking
  4. 1+2+3 > phone does not work
  5. Cause: playing with button > Effect: screen goes white (probability: 10%)
  6. Correlation between playing with button and screen goes white (probability: 75%)

While customer might explain such sequence in different languages and in different words still must discover the right sequence and map all the similar case and events to each other. We expect the event discovery system provides the following capabilities:

  • Find frequent patterns across all customers and across all clients
  • Generalized the discovered model to be exploited for other cases
  • Find correlations between the customer identified event sequence and recommended action by agent
  • Discover the real-time common trends across all call centers
  • Validate the discovered knowledge against a crafted database
  • Classify conversations based on the discovered events
  • Develop an interface for agent to browse, monitor and analyze discovered knowledge
  • Identify potential cause and effect relationships between events
  • Build a knowledge base of all events and normalize the event database
  • Build a knowledge base of all type of discovered relationships
  • Build a knowledge base of all discovered patterns



Applicant should have Master’s Degree or PhD in Computer Science, Electrical Engineering, Statistics or Math with focus on speech processing, statistical knowledge and proficiency with a statistical analysis tools like R or SAS. Should have experience with various text mining and speech processing tools and packages, experience with speech processing, acoustic modelling, language modelling, speech recognition, deep learning algorithm development for speech and text processing.

Should have experience with distributed databases and query languages like SQL, CQL or Pig and programming experience with scripting languages such as Phyton or Ruby.

The applicant must be fluent in statistical analysis and machine learning fundamentals such as regressions, decision trees, neural networks, deep learning, LDA and SVM. Must have experience with machine learning tools and open source packages (e.g. Mallet, TensorFlow, Caffe, Torch7, Theano) and Big Data Processing (Hadoop, Spark, Flink, Kafka)

The potential applicants must meet the conditions set out within Marie Sklodowska-Curie Actions Individual Fellowships call