Using interaction data for improving the offline and online evaluation of search engines

This thesis investigates how web search evaluation can be improved using historical interaction data. Modern search engines combine offline and online evaluation approaches in a sequence of steps that a tested change needs to pass through to be accepted as an improvement and subsequently deployed. We refer to such a sequence of steps as an evaluation pipeline. In this thesis, we consider the evaluation pipeline to contain three sequential steps: an offline evaluation step, an online evaluation scheduling step, and an online evaluation step. In this thesis we show that historical user interaction data can aid in improving the accuracy or efficiency of each of the steps of the web search evaluation pipeline. As a result of these improvements, the overall efficiency of the entire evaluation pipeline is increased. Firstly, we investigate how user interaction data can be used to build accurate offline evaluation methods for query auto-completion mechanisms. We propose a family of offline evaluation metrics for query auto-completion that represents the effort the user has to spend in order to submit their query. The parameters of our proposed metrics are trained against a set of user interactions recorded in the search engine’s query logs. From our experimental study, we observe that our proposed metrics are significantly more correlated with an online user satisfaction indicator than the metrics proposed in the existing literature. Hence, fewer changes will pass the offline evaluation step to be rejected after the online evaluation step. As a result, this would allow us to achieve a higher efficiency of the entire evaluation pipeline. Secondly, we state the problem of the optimised scheduling of online experiments. We tackle this problem by considering a greedy scheduler that prioritises the evaluation queue according to the predicted likelihood of success of a particular experiment. This predictor is trained on a set of online experiments, and uses a diverse set of features to represent an online experiment. Our study demonstrates that a higher number of successful experiments per unit of time can be achieved by deploying such a scheduler on the second step of the evaluation pipeline. Consequently, we argue that the efficiency of the evaluation pipeline can be increased. Next, to improve the efficiency of the online evaluation step, we propose the Generalised Team Draft interleaving framework. Generalised Team Draft considers both the interleaving policy (how often a particular combination of results is shown) and click scoring (how important each click is) as parameters in a data-driven optimisation of the interleaving sensitivity. Further, Generalised Team Draft is applicable beyond domains with a list-based representation of results, i.e. in domains with a grid-based representation, such as image search. Our study using datasets of interleaving experiments performed both in document and image search domains demonstrates that Generalised Team Draft achieves the highest sensitivity. A higher sensitivity indicates that the interleaving experiments can be deployed for a shorter period of time or use a smaller sample of users. Importantly, Generalised Team Draft optimises the interleaving parameters w.r.t. historical interaction data recorded in the interleaving experiments. Finally, we propose to apply the sequential testing methods to reduce the mean deployment time for the interleaving experiments. We adapt two sequential tests for the interleaving experimentation. We demonstrate that one can achieve a significant decrease in experiment duration by using such sequential testing methods. The highest efficiency is achieved by the sequential tests that adjust their stopping thresholds using historical interaction data recorded in diagnostic experiments. Our further experimental study demonstrates that cumulative gains in the online experimentation efficiency can be achieved by combining the interleaving sensitivity optimisation approaches, including Generalised Team Draft, and the sequential testing approaches. Overall, the central contributions of this thesis are the proposed approaches to improve the accuracy or efficiency of the steps of the evaluation pipeline: the offline evaluation frameworks for the query auto-completion, an approach for the optimised scheduling of online experiments, a general framework for the efficient online interleaving evaluation, and a sequential testing approach for the online search evaluation. The experiments in this thesis are based on massive real-life datasets obtained from Yandex, a leading commercial search engine. These experiments demonstrate the potential of the proposed approaches to improve the efficiency of the evaluation pipeline.

Methods matter: computational modelling in public health policy and planning

This work is aimed at understanding and unifying information on epidemiological modelling methods and how those methods relate to public policy addressing human health, specifically in the context of infectious disease prevention, pandemic planning, and health behaviour change. This thesis employs multiple qualitative and quantitative methods, and presents as a manuscript of several individual, data-driven projects that are combined in a narrative arc. The first chapter introduces the scope and complexity of this interdisciplinary undertaking, describing several topical intersections of importance. The second chapter begins the presentation of original data, and describes in detail two exercises in computational epidemiological modelling pertinent to pandemic influenza planning and policy, and progresses in the next chapter to present additional original data on how the confidence of the public in modelling methodology may have an effect on their planned health behaviour change as recommended in public health policy. The thesis narrative continues in the final data-driven chapter to describe how health policymakers use modelling methods and scientific evidence to inform and construct health policies for the prevention of infectious diseases, and concludes with a narrative chapter that evaluates the breadth of this data and recommends strategies for the optimal use of modelling methodologies when informing public health policy in applied public health scenarios.

Exploring coordination-driven self-assembly with autonomous chemical robots

The work presented herein focused on the automation of coordination-driven self assembly, exploring methods that allow syntheses to be followed more closely while forming new ligands, as part of the fundamental study of the digitization of chemical synthesis and discovery. Whilst the control and understanding of the principle of pre-organization and self-sorting under non-equilibrium conditions remains a key goal, a clear gap has been identified in the absence of approaches that can permit fast screening and real-time observation of the reaction process under different conditions. A firm emphasis was thus placed on the realization of an autonomous chemical robot, which can not only monitor and manipulate coordination chemistry in real-time, but can also allow the exploration of a large chemical parameter space defined by the ligand building blocks and the metal to coordinate. The self-assembly of imine ligands with copper and nickel cations has been studied in a multi-step approach using a self-built flow system capable of automatically controlling the liquid-handling and collecting data in real-time using a benchtop MS and NMR spectrometer. This study led to the identification of a transient Cu(I) species in situ which allows for the formation of dimeric and trimeric carbonato bridged Cu(II) assemblies. Furthermore, new Ni(II) complexes and more remarkably also a new binuclear Cu(I) complex, which usually requires long and laborious inert conditions, could be isolated. The study was then expanded to the autonomous optimization of the ligand synthesis by enabling feedback control on the chemical system via benchtop NMR. The synthesis of new polydentate ligands has emerged as a result of the study aiming to enhance the complexity of the chemical system to accelerate the discovery of new complexes. This type of ligand consists of 1-pyridinyl-4-imino-1,2,3-triazole units, which can coordinate with different metal salts. The studies to test for the CuAAC synthesis via microwave lead to the discovery of four new Cu complexes, one of them being a coordination polymer obtained from a solvent dependent crystallization technique. With the goal of easier integration into an automated system, copper tubing has been exploited as the chemical reactor for the synthesis of this ligand, as it efficiently enhances the rate of the triazole formation and consequently promotes the formation of the full ligand in high yields within two hours. Lastly, the digitization of coordination-driven self-assembly has been realized for the first time using an in-house autonomous chemical robot, herein named the ‘Finder’. The chemical parameter space to explore was defined by the selection of six variables, which consist of the ligand precursors necessary to form complex ligands (aldehydes, alkineamines and azides), of the metal salt solutions and of other reaction parameters – duration, temperature and reagent volumes. The platform was assembled using rounded bottom flasks, flow syringe pumps, copper tubing, as an active reactor, and in-line analytics – a pH meter probe, a UV-vis flow cell and a benchtop MS. The control over the system was then obtained with an algorithm capable of autonomously focusing the experiments on the most reactive region (by avoiding areas of low interest) of the chemical parameter space to explore. This study led to interesting observations, such as metal exchange phenomena, and also to the autonomous discovery of self assembled structures in solution and solid state – such as 1-pyridinyl-4-imino-1,2,3-triazole based Fe complexes and two helicates based on the same ligand coordination motif.