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.

Supporting collocated and at-a-distance experiences with TV and VR displays

Televisions (TVs) and VR Head-Mounted Displays (VR HMDs) are used in shared and social spaces in the home. This thesis posits that these displays do not sufficiently reflect the collocated, social contexts in which they reside, nor do they sufficiently support shared experiences at-a-distance. This thesis explores how the role of TVs and VR HMDs can go beyond presenting a single entertainment experience, instead supporting social and shared use in both collocated and at-a-distance contexts. For collocated TV, this thesis demonstrates that the TV can be augmented to facilitate multi-user interaction, support shared and independent activities and multi-user use through multi-view display technology, and provide awareness of the multi-screen activity of those in the room, allowing the TV to reflect the social context in which it resides. For at-a-distance TV, existing smart TVs are shown to be capable of supporting synchronous at-a-distance activity, broadening the scope of media consumption beyond the four walls of the home. For VR HMDs, collocated proximate persons can be seamlessly brought into mixed reality VR experiences based on engagement, improving VR HMD usability. Applied to at-a-distance interactions, these shared mixed reality VR experiences can enable more immersive social experiences that approximate viewing together as if in person, compared to at-a-distance TV. Through an examination of TVs and VR HMDs, this thesis demonstrates that consumer display technology can better support users to interact, and share experiences and activities, with those they are close to.

Interaction techniques with novel multimodal feedback for addressing gesture-sensing systems

Users need to be able to address in-air gesture systems, which means finding where to perform gestures and how to direct them towards the intended system. This is necessary for input to be sensed correctly and without unintentionally affecting other systems. This thesis investigates novel interaction techniques which allow users to address gesture systems properly, helping them find where and how to gesture. It also investigates audio, tactile and interactive light displays for multimodal gesture feedback; these can be used by gesture systems with limited output capabilities (like mobile phones and small household controls), allowing the interaction techniques to be used by a variety of device types. It investigates tactile and interactive light displays in greater detail, as these are not as well understood as audio displays. Experiments 1 and 2 explored tactile feedback for gesture systems, comparing an ultrasound haptic display to wearable tactile displays at different body locations and investigating feedback designs. These experiments found that tactile feedback improves the user experience of gesturing by reassuring users that their movements are being sensed. Experiment 3 investigated interactive light displays for gesture systems, finding this novel display type effective for giving feedback and presenting information. It also found that interactive light feedback is enhanced by audio and tactile feedback. These feedback modalities were then used alongside audio feedback in two interaction techniques for addressing gesture systems: sensor strength feedback and rhythmic gestures. Sensor strength feedback is multimodal feedback that tells users how well they can be sensed, encouraging them to find where to gesture through active exploration. Experiment 4 found that they can do this with 51mm accuracy, with combinations of audio and interactive light feedback leading to the best performance. Rhythmic gestures are continuously repeated gesture movements which can be used to direct input. Experiment 5 investigated the usability of this technique, finding that users can match rhythmic gestures well and with ease. Finally, these interaction techniques were combined, resulting in a new single interaction for addressing gesture systems. Using this interaction, users could direct their input with rhythmic gestures while using the sensor strength feedback to find a good location for addressing the system. Experiment 6 studied the effectiveness and usability of this technique, as well as the design space for combining the two types of feedback. It found that this interaction was successful, with users matching 99.9% of rhythmic gestures, with 80mm accuracy from target points. The findings show that gesture systems could successfully use this interaction technique to allow users to address them. Novel design recommendations for using rhythmic gestures and sensor strength feedback were created, informed by the experiment findings.

Utilising presence in places to support mobile interaction

Physical places are given contextual meaning by the objects and people that make up the space. Presence in physical places can be utilised to support mobile interaction by making access to media and notifications on a smartphone easier and more visible to other people. Smartphone interfaces can be extended into the physical world in a meaningful way by anchoring digital content to artefacts, and interactions situated around physical artefacts can provide contextual meaning to private manipulations with a mobile device. Additionally, places themselves are designed to support a set of tasks, and the logical structure of places can be used to organise content on the smartphone. Menus that adapt the functionality of a smartphone can support the user by presenting the tools most likely to be needed just-in-time, so that information needs can be satisfied quickly and with little cognitive effort. Furthermore, places are often shared with people whom the user knows, and the smartphone can facilitate social situations by providing access to content that stimulates conversation. However, the smartphone can disrupt a collaborative environment, by alerting the user with unimportant notifications, or sucking the user in to the digital world with attractive content that is only shown on a private screen. Sharing smartphone content on a situated display creates an inclusive and unobtrusive user experience, and can increase focus on a primary task by allowing content to be read at a glance. Mobile interaction situated around artefacts of personal places is investigated as a way to support users to access content from their smartphone while managing their physical presence. A menu that adapts to personal places is evaluated to reduce the time and effort of app navigation, and coordinating smartphone content on a situated display is found to support social engagement and the negotiation of notifications. Improving the sensing of smartphone users in places is a challenge that is out-with the scope of this thesis. Instead, interaction designers and developers should be provided with low-cost positioning tools that utilise presence in places, and enable quantitative and qualitative data to be collected in user evaluations. Two lightweight positioning tools are developed with the low-cost sensors that are currently available: The Microsoft Kinect depth sensor allows movements of a smartphone user to be tracked in a limited area of a place, and Bluetooth beacons enable the larger context of a place to be detected. Positioning experiments with each sensor are performed to highlight the capabilities and limitations of current sensing techniques for designing interactions with a smartphone. Both tools enable prototypes to be built with a rapid prototyping approach, and mobile interactions can be tested with more advanced sensing techniques as they become available. Sensing technologies are becoming pervasive, and it will soon be possible to perform reliable place detection in-the-wild. Novel interactions that utilise presence in places can support smartphone users by making access to useful functionality easy and more visible to the people who matter most in everyday life.

Machine learning techniques for implicit interaction using mobile sensors

Interactions in mobile devices normally happen in an explicit manner, which means that they are initiated by the users. Yet, users are typically unaware that they also interact implicitly with their devices. For instance, our hand pose changes naturally when we type text messages. Whilst the touchscreen captures finger touches, hand movements during this interaction however are unused. If this implicit hand movement is observed, it can be used as additional information to support or to enhance the users’ text entry experience. This thesis investigates how implicit sensing can be used to improve existing, standard interaction technique qualities. In particular, this thesis looks into enhancing front-of-device interaction through back-of-device and hand movement implicit sensing. We propose the investigation through machine learning techniques. We look into problems on how sensor data via implicit sensing can be used to predict a certain aspect of an interaction. For instance, one of the questions that this thesis attempts to answer is whether hand movement during a touch targeting task correlates with the touch position. This is a complex relationship to understand but can be best explained through machine learning. Using machine learning as a tool, such correlation can be measured, quantified, understood and used to make predictions on future touch position. Furthermore, this thesis also evaluates the predictive power of the sensor data. We show this through a number of studies. In Chapter 5 we show that probabilistic modelling of sensor inputs and recorded touch locations can be used to predict the general area of future touches on touchscreen. In Chapter 7, using SVM classifiers, we show that data from implicit sensing from general mobile interactions is user-specific. This can be used to identify users implicitly. In Chapter 6, we also show that touch interaction errors can be detected from sensor data. In our experiment, we show that there are sufficient distinguishable patterns between normal interaction signals and signals that are strongly correlated with interaction error. In all studies, we show that performance gain can be achieved by combining sensor inputs.