Seminar Series



Past Seminars 

Monday 29 April 2019
Automated Text Analysis to Detect Rare Social Problems


Speaker: Leke de Vries (Doctoral Candidate & National Institute of Justice Graduate Research Fellow, Northeastern University)

Date: Monday 29 April 2019, 12.30pm - 2.00pm

Room: Department of Methodology, Columbia House, LSE (COL.8.13)

Abstract: This talk presents how automated text analyses help illuminate relatively rare phenomena in large amounts of text. Human trafficking victimizations will be utilized as a case to illustrate the use of mathematical text analysis to reveal crime patterns and suggest new directions for research on crime or other complex problems. As has been noted in recent work, the application of supervised methods to detect and understand crime is challenged by lacking or biased ground truth. The examination of various crime patterns requires an inductive approach to the use of machine-learning techniques. The talk highlights the use of computational techniques based on distributional word and text representations to present indicators of crime-related activities. Distributional word and text representations have been used to detect other types of crimes for which limited or biased ground truth exists, including hate crime, abusive language, or gender biases. The presentation will focus on how patterns for rather infrequent words can be understood through models that weigh nearby-context words more heavily than distant words and suggests the use of qualitative methods to guide the model specification and interpretation.

Thursday 28 March 2019
Personalized Dynamic Pricing with Machine Learning


Gah Yi

Speaker: Dr Gah-Yi Ban (Assistant Professor of Management Science and Operations, London Business School)

Date: Thursday 28 March 2019, 2.30pm - 4.00pm

Room: Leverhulme Library, Columbia House, LSE (COL.6.15)

Abstract: We consider a seller who can dynamically adjust the price of a product at the individual customer level, by utilizing information about customers’ characteristics encoded as a d-dimensional feature vector. We assume a personalized demand model, parameters of which depend on s out of the d features. The seller initially does not know the relationship between the customer features and the product demand, but learns this through sales observations over a selling horizon of T periods. We prove that the seller’s expected regret, i.e., the revenue loss against a clairvoyant who knows the underlying demand relationship, is at least of order s √T under any admissible policy. We then design a near-optimal pricing policy for a “semi-clairvoyant” seller (who knows which s of the d features are in the demand model) that achieves an expected regret of order s √Tlog T. We extend this policy to a more realistic setting where the seller does not know the true demand predictors, and show this policy has an expected regret of order s √T(log d+logT), which is also near-optimal. Finally, we test our theory on simulated data and on a data set from an online auto loan company in the United States. On both data sets, our experimentation-based pricing policy is superior to intuitive and/or widely-practiced customized pricing methods such as myopic pricing and segment-then-optimize policies. Furthermore, our policy improves upon the loan company’s historical pricing decisions by 32% in terms of annual expected revenue.


Wednesday 27 March 2019
An improved method for fitting item response theory models to sparse data



Speaker: Kevin Quinn (Professor, Department of Political Science, Univeristy of Michigan

Date: Wednesday 27 March 2019, 12.30pm - 2.00pm

Room: Department of Methodology, Columbia House, LSE (COL.8.13)

Abstract: Item response theory (IRT) models are widely used by political scientists to measure latent concepts such as attitudes or ideology. Unlike the canonical psychometric applications, many applications in political science feature extremely sparse data as well as a lack of prior information about the sign of the discrimination parameters. This lack of information can result in a posterior density with multiple modes—even when conventional identification assumptions are employed. In such situations, standard model-fitting strategies based on Gibbs sampling and data augmentation do a poor job of exploring all high probability areas of the parameter space. We develop a solution to this problem based on the idea of parallel tempering. We illustrate the approach with examples drawn from political science. 


Thursday 14 March 2019
Peer influence and the spreading of music at Spotify


Speaker: Dr. Marc Kueschnigg (Associate Professor/Deputy Director, Institute for Analytical Sociology, Linköping University)

Date: Thursday 14 March 2019, 2.30pm - 4.00pm

Room: Leverhulme Library, Columbia House, LSE (COL.6.15)

Abstract: The wide availability of online data marks a new era of investigation into human behavior embedded in social environments, and it allows us to change the focus of the analysis from that of the individual-level correlates of human behavior to the dynamics of populations of interacting individuals. Scrutinizing cultural choice at, a leading online music platform, we identify peer effects of social contagion among interconnected consumers of music to gain understanding of the mechanisms underlying the emergence of hits, the establishment of new artists and genres, and cultural change more generally. We utilize a massive networked dataset on music preferences and listening patterns collected through the Spotify API to estimate peer influence in the adoption of novel music. A granular measurement of music taste permits high-dimensional matching of users who either have or have not been exposed to new music through their network contacts. To overcome the selection bias typical for observational studies relying on statistical matching, we estimate treatment effects considering thousands of individual and contextual pre-treatment covariates which we measure and pre-process through unsupervised machine learning. This novel approach permits us to consider innovation characteristics, status differences between influencers and imitators, and treatment dosages moderating the strength of peer influence. Most importantly, our design allows us to determine whether peer influence merely informs people about current trends or whether it is truly persuasive---i.e. forceful enough to change people’s behavior and thus bring about surprising aggregate outcomes that do not map the distribution of individual preferences in a population but lead to the diffusion of the “unexpectable” resulting from path-dependent processes that could have produced a very different reality under similar circumstances.

Thursday 28 February 2019
Living with intelligent machines


Speaker: Nello Cristianini (Professor of Artificial Intelligence, University of Bristol)

Date: Thursday 28 February 2019, 2.30pm - 4.00pm

Room: Leverhulme Library, Columbia House, LSE (COL.6.15)

Abstract: The way we build AI agents is based on machine learning and vast masses of training data, which is often the result of human activities.The consequences of this methodology are manifold, from privacy concerns to issues relative to implicit bias and discrimination. The business model underlying much of AI today is based on personalisation of services, which - for some - can generate concerns of personal data collection, possible manipulation of behaviour, and unexplored questions about dynamic pricing, opinion polarisation and possibly addiction. Understanding the technical and business models behind modern AI helps us understand the various challenges we face, devise regulation, and also consider possible negative effects of new technologies once deployed in society. 

Thursday 7 February 2019
Matching with text data: An experimental evaluation of methods for matching documents and of measuring match quality


Speaker: Aaron R. Kaufman (PhD Candidate, Harvard University)

Date: Thursday 7 February 2019, 2.30pm - 4.00pm

Room: Leverhulme Library, Columbia House, LSE (COL.6.15)

Abstract: Matching for causal inference is a well-studied problem, but standard methods fail when the units to match are text documents: the high-dimensional and rich nature of the data renders exact matching infeasible, causes propensity scores to produce incomparable matches, and makes assessing match quality difficult. In this talk, Aaron Kaufman characterizes a framework for matching text documents that decomposes existing methods into: (1) the choice of text representation, and (2) the choice of distance metric, investigate how different choices within this framework affect both the quantity and quality of matches identified through a systematic multifactor evaluation experiment using human subjects, and apply the results to the study of media bias.

Thursday 24 January 2019
Unfolding-Model-Based visualization: Theory, method and application


Speaker: Dr Yunxiao Chen (Assistant Professor, Department of Statistics, LSE)

Date: Thursday 7 February 2019, 2.30pm - 4.00pm

Room: Leverhulme Library, Columbia House, LSE (COL.6.15)

Abstract: Multidimensional unfolding methods are widely used for visualizing item response data. Such methods project respondents and items simultaneously onto a low-dimensional space, in which respondents and items are represented by ideal points with person-person, item-item, and person-item similarities being captured by the Euclidian distances between the points. In this paper, we study the visualization of multidimensional unfolding from a statistical perspective. We cast multidimensional unfolding into an estimation problem, where the respondent and item ideal points are treated as parameters to estimate. An estimator is then proposed for the simultaneous estimation of these parameters. An asymptotic theory and non-asymptotic error bounds are provided for the recovery of the ideal points, shedding lights on the validity of model-based visualization. An alternating projected gradient descent algorithm is proposed for parameter estimation. We provide two illustrative examples of users' movie rating and of senate roll call voting. (This is a joint project with Dr. Zhiliang Ying and Mr. Haoran Zhang).

Thursday 13 December 2018
Pursuing the UN data revolution for sustainable development


Speaker: Dr Viktoria Spaiser (University Academic Fellow in Political Science Informatics, University of Leeds)

Date: Thursday 13 December 2018, 4.15pm - 5.45pm

Room: Leverhulme Library, Columbia House, LSE (COL.6.15)

Abstract: In August 2014 the UN established an Independent Expert Advisory Group to make concrete recommendations on bringing about a data revolution in sustainable development. The hope has been that data analytics would help to deal with the enormous challenge of achieving a sustainable development globally. But what does existing data actually tell us about the challenge and potential solutions? And what other data do we need in order to understand the problem in all its dimensions? Dr Viktoria Spaiser will focus on the global and the individual level of the sustainability challenge. She will discuss recent studies that she conducted with colleagues modelling the compatibility of the UN Sustainable Development Goals on the one hand and studying environmental behaviour in a field-experimental setup on the other hand. Looking globally into the empirically measurable conflict of various Sustainable Development Goals, Dr Spaiser will explain what cross-country time-series data tells us about the nature of the inconsistencies. In this context she will also discuss a recent extension of the original study, examining the different conclusions that can be drawn about the sustainability challenge depending on how the Sustainable Development Goals are operationalized. Specifically, she will show why it does matter whether we look at production-based or consumption-based CO2 emissions when pursuing the Sustainable Development Agenda. Dr Spaiser will then change the perspective and look at the sustainability challenge from an individual level angle. The core question here is: how can we encourage more pro-environmental behaviour? She will discuss a pilot study conducted recently, making use of smartphones to collect daily environmental behaviour data in a field-experimental setup. Dr Spaiser will conclude with a programmatic note on the road ahead in the sustainability research she is envisioning.

Wednesday 28 November 2018
Exposure to opposing views on social media can increase political polarization


Speaker: Professor Chris Bail (Professor of Sociology and Public Policy, Duke University)

Date: Wednesday 28 November 2018, 12.30pm - 2.00pm

Room: Department of Methodology, Columbia House, LSE (COL.8.13)

Abstract: There is mounting concern that social media sites contribute to political polarization by creating “echo chambers” that insulate people from opposing views about current events. We surveyed a large sample of Democrats and Republicans who visit Twitter at least three times each week about a range of social policy issues. One week later, we randomly assigned respondents to a treatment condition in which they were offered financial incentives to follow a Twitter bot for 1 month that exposed them to messages from those with opposing political ideologies (e.g., elected officials, opinion leaders, media organizations, and nonprofit groups). Respondents were resurveyed at the end of the month to measure the effect of this treatment, and at regular intervals throughout the study period to monitor treatment compliance. We find that Republicans who followed a liberal Twitter bot became substantially more conservative posttreatment. Democrats exhibited slight increases in liberal attitudes after following a conservative Twitter bot, although these effects are not statistically significant. Notwithstanding important limitations of our study, these findings have significant implications for the interdisciplinary literature on political polarization and the emerging field of computational social science.


Thursday 15 November 2018
Combining forecastes in the presence of ambiguity over correlation structures


Speaker: Professor Gilat Levy & Professor Ronny Razin (Department of Economics, LSE)

Date: Thursday 15 November 2018, 4.15pm - 5.45pm

Room: Leverhulme Library, Columbia House, LSE (COL.6.15)


Thursday 25 October 2018
Tourism and terrorism: The impact of news reporting


Speaker: Professor Sir Tim Besley (Department of Economics, LSE)

Date: Thursday 15 November 2018, 4.15pm - 5.45pm

Room: Leverhulme Library, Columbia House, LSE (COL.6.15)