View on GitHub

Bennett Kleinberg

I am an Assistant Professor in Data Science at the Department of Security and Crime Science and the Dawes Centre for Future Crime at University College London. I also hold a guest affiliation at the University of Amsterdam’s Department of Psychology.

News

6/2020 - Paper on experimental emotion manipulation for computational linguistics: https://arxiv.org/abs/2006.08952
5/2020 - Media coverage (in French) of our drill music work on Ulyces
5/2020 - Media coverage of our COVID-19 study in Daily Mail and UCL News
4/2020 - Paper on gender differences in emotional responses to COVID-19: http://arxiv.org/abs/2004.08202.
4/2020 - Media coverage of our drill music work in Discover Magazine
4/2020 - Paper on detecting adversarial examples in NLP (PhD work by M. Mozes): http://arxiv.org/abs/2004.05887.
4/2020 - Follow-up paper on the drill rap - violence issue: http://arxiv.org/abs/2004.04598.
4/2020 - New paper introducing a high-quality dataset on emotions around COVID-19: http://arxiv.org/abs/2004.04225
3/2020 - Paper on human-machine integrated deception detection: http://arxiv.org/abs/2003.13316.
2/2020 - Paper on the temporal evolution of a far-right forum in the Journal of Computational Social Science
1/2020 - Paper on using a smartphone app for RT-based deception detection in Collabra.
1/2020 - Paper on human trafficking in online ads in Trends in Organized Crime.

My research currently falls into two areas:

(1) Computational approaches to security and crime problems

With more data and a changing landscape of problems, computational techniques are an exciting way to look at existing and novel crime and security problems. I’m especially interested in emerging crime problems and understanding them through complex text data.

Techniques used: natural language processing, applied machine learning.

Ideas I’m working on:

examining new (crime) problems on the Internet through text (e.g., online forums)
dynamic language use (e.g. linguistic trajectories within persons/texts/forums)
keyboard dynamics as a biometric identifier

(2) Methods of behavioural data science

Before computational methods can help us understand problems better (or at scale), we need suitable methods and tested assumptions. I’m particularly interested in building robust methods to learn about human behaviour from text and web data. Techniques: behavioural experimental research, statistical modelling, text mining.

Ideas I’m working on:

examining the fundamental assumptions of behavioural data science (e.g. the cognition-language-behaviour nexus)
inferring behaviour and cognition from text data
automated, learning-based text anonymisation (SAGE Ocean concept grant winner), (more info), (interviews on YouTube)
sample size estimation for machine learning in the social and behavioural sciences

For recent talks/workshops, see here

Some of the ongoing and past projects were funded by SAGE, RUSI/GIFCT, The Dutch Ministry of Security and Justice

Media

Ulyces (French), UCL News II, Daily Mail, Disover Magazine, YouTube SAGE, New Scientist II, BBC Newsroom, The Times, UCL News I, The Guardian, Deutsche Welle (German), Süddeutsche Zeitung (German), Volkskrant II (Dutch), Folia Magazine (Dutch), New Scientist I, BNR Newsradio, Volkskrant I (Dutch), Amsterdam FM (Dutch), Psychology Today, APA News

Research group

PhD students

Isabelle van der Vegt (website)

Background: Computational Linguistics, Psychology
Topic: Understanding and predicting threats of violence using computational linguistics
Supervised with Prof Paul Gill
Funded by Prof P Gill’s ERC grant ‘GRIEVANCE’

Felix Soldner (website)

Background: Psychology, Brain & Cognitive Sciences
Topic: Detecting and mitigating online consumer fraud
Supervised with Prof Shane Johnson
Funded by: UCL Dawes Centre for Future Crime

Maximilian Mozes (website)

Background: Computer Science, Natural Language Processing
Topic: Adversarial perturbations in natural language processing
Supervised with Prof Lewis Griffin
Funded by: UCL Dawes Centre for Future Crime

Daniel Hammocks (website)

Background: Mathematics, Data Science
Topic: Detecting emerging crimes using data science techniques
Supervised with Prof Kate Bowers
Funded by: EPSRC

Josh Kamps

Background: Computer Science, Crime Science
Topic: Detecting cryptocurrency crime
Supervised with Dr Sarah Meiklejohn
Funded by: UCL Dawes Centre for Future Crime

Arianna Trozze

Background: Law, Policy Making
Topic: Detecting cryptocurrency crime
Supervised with Dr Toby Davies
Funded by: UCL Cybercrime Centre for Doctoral Training

Sergi Bray

Background: Computer Science, Machine Learning
Topic: Behavioural and cognitive aspects of deep fakes perception
Supervised with Prof Shane Johnson
Funded by: UCL Cybercrime Centre for Doctoral Training

Dawes reading group

We are running a bi-weekly reading group on topics around future crime (incl. data science, policy-making). Info at https://felix-soldner.github.io/dawes_reading_group/

Dissertation projects

Past dissertation projects

Teaching

2019/2020
- SECU0050, Data Science for Crime Scientists (BSc), UCL, Material on GitHub
- SECU0057, Applied Data Science (MSc), UCL, Material on GitHub
2018/2019
- SECU0013, Probability, Statistics & Modelling 2 (BSc), UCL, Material on GitHub
- SECU0050, Data Science for Crime Scientists (BSc), UCL, Material on GitHub
- Crime Science (BSc honours module), University of Amsterdam (study guide)
2017/2018
- Crime Science (open UvA module), University of Amsterdam
- Crime Science (BSc honours module), University of Amsterdam (study guide)
- Deception and Lie Detection (MSc), University of Amsterdam (study guide)
2016/2017
- Deception and Lie Detection (MSc), University of Amsterdam (study guide)
2015/2016
- Deception and Lie Detection (MSc), University of Amsterdam

Software

NETANOS - Named entity-based text anonymization for open science
- Anonymises bunches of text files by removing identifiable information
- de-identifies numbers, persons, locations, places, pronouns, dates and times
- Developed with Maximilian Mozes and Bruno Verschuere
- Available on npm, source code on GitHub
- Published paper in JOSS and experimental validation on OSF servers
- Accessible interface version lives on netanos.io
Naive context sentiment analysis
- Performs sentiment analysis that handles valence shifters (e.g., “really”, “not”, “hardly”, “but”) on non-punctuated or poorly punctuated data
- Code lives on GitHub
- Specific use cases:
  - data that are not punctuated at all (e.g., auto-generated YouTube transcripts)
  - data that are badly punctuated (e.g., data from blogs where punctuation is not necessarily a given)
  - very brief text data: Twitter data, for example, even if properly annotated for sentence-boundary-disambiguation, would return one or two sentiment values with other sentiment extraction packages
- Contributors: Maximilian Mozes and Isabelle van der Vegt

Brief CV

Assistant Professor in Data Science (Department of Security and Crime Science, UCL)
PhD on detecting deceptive intentions on a large scale (Department of Psychology, University of Amsterdam) [2018]
MSc in Crime Science (University College London) [2015]
BSc in Psychology - specialised in Psychological Methods (University of Amsterdam) [2014]

Links:

Review service

Scientific Reports (Nature); Psychological Science; Computers in Human Behavior; Plos One; Perspectives on Psychological Science; Affective Computing & Intelligent Interaction conference proceedings; Information; ACL 2020; AACL-IJCNLP 2020; Memory & Cognition; Applied Cognitive Psychology; Crime Science; Acta Psychologica; Personality and Individual Differences; Current Psychology; Cognitive Research: Principles and Implications; EuroCSS 2019; Computers & Security; Psychology, Crime & Law; Legal & Criminological Psychology.

Publications

All publications can be found on my Google Scholar page.

A list updated in April 2020 is available here

Contact

If you are interested in doing a PhD, working as a research assistant (in London or remotely) or doing a BSc/MSc dissertation, feel free to reach out to me.