Pie & AI: Hamburg - Natural Language Processing Meetup

Pie & AI is a series of DeepLearning.AI meetups independently hosted by community groups. This event is hosted by Natural Language Processing Hamburg.

Trailer
https://youtu.be/Qz3ZVYuQUEM

Please register with Eventbrite:
https://www.eventbrite.co.uk/e/pie-ai-hamburg-natural-language-processing-meetup-tickets-138981496397

Event Agenda & Speakers:

18:30 - 18:45 Virtual Doors Open + Networking

18:45 - 19:15 Introduction by the organizers and general remarks

19:15 - 19:45 "Training a sentiment model"
Ankit Singh (Data Scientist – Ex-Cauliflower)

19:45 - 20:00 Questions, Discussion and Coffee Break

20:00 - 20:30 "Deception detection in the conversation using Linguistic Markers"
Nikesh Bajaj (Postdoctoral Research Fellow at University of East London)

20:30 ~ 21:00 Questions, Discussion and open end

This event will be held as a Pie & AI Community Event:
https://www.eventbrite.com/e/pie-ai-hamburg-natural-language-processing-meetup-tickets-138981496397

Abstracts:

Deception detection in the conversation using Linguistic Markers / Nikesh Bajaj

Abstract: Detecting the elements of deception in a conversation with automated system is a challenging task for AI community. In a limited available literature, studies have used approach to count the psycholinguistic markers of different categories, such as Negation, Uncertainty, Cognitive etc, to evaluate the deception probability. However, such approach loses the interaction between different markers. In collaboration behaviour experts, we introduce an approach that exploits the proximity of linguistic markers to each other. Using this we encapsulate the interaction of markers such as their impact on each other with respect to proximity. The proposed approach is used to design the Decision Engine, that produce a deception score (likelihood) for a speaker in a conversation. The approach has been evaluated on two datasets, namely; (1) Columbia-SRI-Colorado (CSC) deceptive speech dataset and (2) a real-world Financial Services dataset, achieved testing accuracy of 69% and 72% respectively. The linguistic analysis, conducted on trained model reveals the interesting characteristics of deceptive behaviour, which are validated by behaviour experts. The work is progressing to include additional acoustic features to make system robust and scalable across different context.

___________
For learners of all levels; beginners, intermediate and advanced