arXiv subjects

Julia Kiseleva

Publications and source records attributed to Julia Kiseleva.

34 records · Page 2Linked to original sources

arXiv2021

Improving Response Quality with Backward Reasoning in Open-domain Dialogue Systems

Being able to generate informative and coherent dialogue responses is crucial when designing human-like open-domain dialogue systems. Encoder-decoder-based dialogue models tend to produce generic and dull responses during the decoding step because the most predictable response is likely to be a non-informative response instead of the most suitable one. To alleviate this problem, we propose to train the generation model in a bidirectional manner by adding a backward reasoning step to the vanilla encoder-decoder training. The proposed backward reasoning step pushes the model to produce more informative and coherent content because the forward generation step's output is used to infer the dialogue context in the backward direction. The advantage of our method is that the forward generation and backward reasoning steps are trained simultaneously through the use of a latent variable to facilitate bidirectional optimization. Our method can improve response quality without introducing side information (e.g., a pre-trained topic model). The proposed bidirectional response generation method achieves state-of-the-art performance for response quality.

cs.CL↗

arXiv2021

How to Not Measure Disentanglement

To evaluate disentangled representations several metrics have been proposed. However, theoretical guarantees for conventional metrics of disentanglement are missing. Moreover, conventional metrics do not have a consistent correlation with the outcomes of qualitative studies. In this paper we analyze metrics of disentanglement and their properties. We conclude that existing metrics of disentanglement were created to reflect different characteristics of disentanglement and do not satisfy two basic desirable properties: (1) assign a high score to representations that are disentangled according to the definition; and (2) assign a low score to representations that are entangled according to the definition. In addition, we propose a new metric of disentanglement and prove that it satisfies both of the properties.

cs.LG↗

arXiv2020

ConvAI3: Generating Clarifying Questions for Open-Domain Dialogue Systems (ClariQ)

This document presents a detailed description of the challenge on clarifying questions for dialogue systems (ClariQ). The challenge is organized as part of the Conversational AI challenge series (ConvAI3) at Search Oriented Conversational AI (SCAI) EMNLP workshop in 2020. The main aim of the conversational systems is to return an appropriate answer in response to the user requests. However, some user requests might be ambiguous. In IR settings such a situation is handled mainly thought the diversification of the search result page. It is however much more challenging in dialogue settings with limited bandwidth. Therefore, in this challenge, we provide a common evaluation framework to evaluate mixed-initiative conversations. Participants are asked to rank clarifying questions in an information-seeking conversations. The challenge is organized in two stages where in Stage 1 we evaluate the submissions in an offline setting and single-turn conversations. Top participants of Stage 1 get the chance to have their model tested by human annotators.

cs.CL↗

arXiv2020

Rethinking Supervised Learning and Reinforcement Learning in Task-Oriented Dialogue Systems

Dialogue policy learning for task-oriented dialogue systems has enjoyed great progress recently mostly through employing reinforcement learning methods. However, these approaches have become very sophisticated. It is time to re-evaluate it. Are we really making progress developing dialogue agents only based on reinforcement learning? We demonstrate how (1)~traditional supervised learning together with (2)~a simulator-free adversarial learning method can be used to achieve performance comparable to state-of-the-art RL-based methods. First, we introduce a simple dialogue action decoder to predict the appropriate actions. Then, the traditional multi-label classification solution for dialogue policy learning is extended by adding dense layers to improve the dialogue agent performance. Finally, we employ the Gumbel-Softmax estimator to alternatively train the dialogue agent and the dialogue reward model without using reinforcement learning. Based on our extensive experimentation, we can conclude the proposed methods can achieve more stable and higher performance with fewer efforts, such as the domain knowledge required to design a user simulator and the intractable parameter tuning in reinforcement learning. Our main goal is not to beat reinforcement learning with supervised learning, but to demonstrate the value of rethinking the role of reinforcement learning and supervised learning in optimizing task-oriented dialogue systems.

cs.CL↗

arXiv2020

Guided Dialog Policy Learning without Adversarial Learning in the Loop

Reinforcement Learning (RL) methods have emerged as a popular choice for training an efficient and effective dialogue policy. However, these methods suffer from sparse and unstable reward signals returned by a user simulator only when a dialogue finishes. Besides, the reward signal is manually designed by human experts, which requires domain knowledge. Recently, a number of adversarial learning methods have been proposed to learn the reward function together with the dialogue policy. However, to alternatively update the dialogue policy and the reward model on the fly, we are limited to policy-gradient-based algorithms, such as REINFORCE and PPO. Moreover, the alternating training of a dialogue agent and the reward model can easily get stuck in local optima or result in mode collapse. To overcome the listed issues, we propose to decompose the adversarial training into two steps. First, we train the discriminator with an auxiliary dialogue generator and then incorporate a derived reward model into a common RL method to guide the dialogue policy learning. This approach is applicable to both on-policy and off-policy RL methods. Based on our extensive experimentation, we can conclude the proposed method: (1) achieves a remarkable task success rate using both on-policy and off-policy RL methods; and (2) has the potential to transfer knowledge from existing domains to a new domain.

cs.AI↗

arXiv2020

Optimizing Interactive Systems via Data-Driven Objectives

Effective optimization is essential for real-world interactive systems to provide a satisfactory user experience in response to changing user behavior. However, it is often challenging to find an objective to optimize for interactive systems (e.g., policy learning in task-oriented dialog systems). Generally, such objectives are manually crafted and rarely capture complex user needs in an accurate manner. We propose an approach that infers the objective directly from observed user interactions. These inferences can be made regardless of prior knowledge and across different types of user behavior. We introduce Interactive System Optimizer (ISO), a novel algorithm that uses these inferred objectives for optimization. Our main contribution is a new general principled approach to optimizing interactive systems using data-driven objectives. We demonstrate the high effectiveness of ISO over several simulations.

cs.AI↗

arXiv2019

Learning Data-Driven Objectives to Optimize Interactive Systems

Effective optimization is essential for interactive systems to provide a satisfactory user experience. However, it is often challenging to find an objective to optimize for. Generally, such objectives are manually crafted and rarely capture complex user needs in an accurate manner. We propose an approach that infers the objective directly from observed user interactions. These inferences can be made regardless of prior knowledge and across different types of user behavior. We introduce interactive system optimization, a novel algorithm that uses these inferred objectives for optimization. Our main contribution is a new general principled approach to optimizing interactive systems using data-driven objectives. We demonstrate the high effectiveness of interactive system optimization over several simulations.

cs.AI↗

arXiv2019

SEntNet: Source-aware Recurrent Entity Network for Dialogue Response Selection

Dialogue response selection is an important part of Task-oriented Dialogue Systems (TDSs); it aims to predict an appropriate response given a dialogue context. Obtaining key information from a complex, long dialogue context is challenging, especially when different sources of information are available, e.g., the user's utterances, the system's responses, and results retrieved from a knowledge base (KB). Previous work ignores the type of information source and merges sources for response selection. However, accounting for the source type may lead to remarkable differences in the quality of response selection. We propose the Source-aware Recurrent Entity Network (SEntNet), which is aware of different information sources for the response selection process. SEntNet achieves this by employing source-specific memories to exploit differences in the usage of words and syntactic structure from different information sources (user, system, and KB). Experimental results show that SEntNet obtains 91.0% accuracy on the Dialog bAbI dataset, outperforming prior work by 4.7%. On the DSTC2 dataset, SEntNet obtains an accuracy of 41.2%, beating source unaware recurrent entity networks by 2.4%.

cs.IR↗

arXiv2018

Dialogue Generation: From Imitation Learning to Inverse Reinforcement Learning

The performance of adversarial dialogue generation models relies on the quality of the reward signal produced by the discriminator. The reward signal from a poor discriminator can be very sparse and unstable, which may lead the generator to fall into a local optimum or to produce nonsense replies. To alleviate the first problem, we first extend a recently proposed adversarial dialogue generation method to an adversarial imitation learning solution. Then, in the framework of adversarial inverse reinforcement learning, we propose a new reward model for dialogue generation that can provide a more accurate and precise reward signal for generator training. We evaluate the performance of the resulting model with automatic metrics and human evaluations in two annotation settings. Our experimental results demonstrate that our model can generate more high-quality responses and achieve higher overall performance than the state-of-the-art.

cs.CL↗

arXiv2017

Towards Learning Reward Functions from User Interactions

In the physical world, people have dynamic preferences, e.g., the same situation can lead to satisfaction for some humans and to frustration for others. Personalization is called for. The same observation holds for online behavior with interactive systems. It is natural to represent the behavior of users who are engaging with interactive systems such as a search engine or a recommender system, as a sequence of actions where each next action depends on the current situation and the user reward of taking a particular action. By and large, current online evaluation metrics for interactive systems such as search engines or recommender systems, are static and do not reflect differences in user behavior. They rarely capture or model the reward experienced by a user while interacting with an interactive system. We argue that knowing a user's reward function is essential for an interactive system as both for learning and evaluation. We propose to learn users' reward functions directly from observed interaction traces. In particular, we present how users' reward functions can be uncovered directly using inverse reinforcement learning techniques. We also show how to incorporate user features into the learning process. Our main contribution is a novel and dynamic approach to restore a user's reward function. We present an analytic approach to this problem and complement it with initial experiments using the interaction logs of a cultural heritage institution that demonstrate the feasibility of the approach by uncovering different reward functions for different user groups.

cs.IR↗

arXiv2017

Modeling Label Ambiguity for Neural List-Wise Learning to Rank

List-wise learning to rank methods are considered to be the state-of-the-art. One of the major problems with these methods is that the ambiguous nature of relevance labels in learning to rank data is ignored. Ambiguity of relevance labels refers to the phenomenon that multiple documents may be assigned the same relevance label for a given query, so that no preference order should be learned for those documents. In this paper we propose a novel sampling technique for computing a list-wise loss that can take into account this ambiguity. We show the effectiveness of the proposed method by training a 3-layer deep neural network. We compare our new loss function to two strong baselines: ListNet and ListMLE. We show that our method generalizes better and significantly outperforms other methods on the validation and test sets.

cs.IR↗

arXiv2017

Evaluating Personal Assistants on Mobile devices

The iPhone was introduced only a decade ago in 2007 but has fundamentally changed the way we interact with online information. Mobile devices differ radically from classic command-based and point-and-click user interfaces, now allowing for gesture-based interaction using fine-grained touch and swipe signals. Due to the rapid growth in the use of voice-controlled intelligent personal assistants on mobile devices, such as Microsoft's Cortana, Google Now, and Apple's Siri, mobile devices have become personal, allowing us to be online all the time, and assist us in any task, both in work and in our daily lives, making context a crucial factor to consider. Mobile usage is now exceeding desktop usage, and is still growing at a rapid rate, yet our main ways of training and evaluating personal assistants are still based on (and framed in) classical desktop interactions, focusing on explicit queries, clicks, and dwell time spent. However, modern user interaction with mobile devices is radically different due to touch screens with a gesture- and voice-based control and the varying context of use, e.g., in a car, by bike, often invalidating the assumptions underlying today's user satisfaction evaluation. There is an urgent need to understand voice- and gesture-based interaction, taking all interaction signals and context into account in appropriate ways. We propose a research agenda for developing methods to evaluate and improve context-aware user satisfaction with mobile interactions using gesture-based signals at scale.

cs.HC↗

arXiv2016

Beyond Movie Recommendations: Solving the Continuous Cold Start Problem in E-commerceRecommendations

Many e-commerce websites use recommender systems or personalized rankers to personalize search results based on their previous interactions. However, a large fraction of users has no prior inter-actions, making it impossible to use collaborative filtering or rely on user history for personalization. Even the most active users mayvisit only a few times a year and may have volatile needs or different personas, making their personal history a sparse and noisy signal at best. This paper investigates how, when we cannot rely on the user history, the large scale availability of other user interactions still allows us to build meaningful profiles from the contextual data and whether such contextual profiles are useful to customize the ranking, exemplified by data from a major online travel agentBooking.com.Our main findings are threefold: First, we characterize the Continuous Cold Start Problem(CoCoS) from the viewpoint of typical e-commerce applications. Second, as explicit situational con-text is not available in typical real world applications, implicit cues from transaction logs used at scale can capture essential features of situational context. Third, contextual user profiles can be created offline, resulting in a set of smaller models compared to a single huge non-contextual model, making contextual ranking available with negligible CPU and memory footprint. Finally we conclude that, in an online A/B test on live users, our contextual ranker in-creased user engagement substantially over a non-contextual base-line, with click-through-rate (CTR) increased by 20%. This clearly demonstrates the value of contextual user profiles in a real world application.

cs.IR↗

arXiv2015

The Impact of Technical Domain Expertise on Search Behavior and Task Outcome

Domain expertise is regarded as one of the key factors impacting search success: experts are known to write more effective queries, to select the right results on the result page, and to find answers satisfying their information needs. Search transaction logs play the crucial role in the result ranking. Yet despite the variety in expertise levels of users, all prior interactions are treated alike, suggesting that weighting in expertise can improve the ranking for informational tasks. The main aim of this paper is to investigate the impact of high levels of technical domain expertise on both search behavior and task outcome. We conduct an online user study with searchers proficient in programming languages. We focus on Java and Javascript, yet we believe that our study and results are applicable for other expertise-sensitive search tasks. The main findings are three-fold: First, we constructed expertise tests that effectively measure technical domain expertise and correlate well with the self-reported expertise. Second, we showed that there is a clear position bias, but technical domain experts were less affected by position bias. Third, we found that general expertise helped finding the correct answers, but the domain experts were more successful as they managed to detect better answers. Our work is using explicit tests to determine user expertise levels, which is an important step toward fully automatic detection of expertise levels based on interaction behavior. A deeper understanding of the impact of expertise on search behavior and task outcome can enable more effective use of expert behavior in search logs - essentially make everyone search as an expert.

cs.IR↗

arXiv2015

The Continuous Cold Start Problem in e-Commerce Recommender Systems

Many e-commerce websites use recommender systems to recommend items to users. When a user or item is new, the system may fail because not enough information is available on this user or item. Various solutions to this `cold-start problem' have been proposed in the literature. However, many real-life e-commerce applications suffer from an aggravated, recurring version of cold-start even for known users or items, since many users visit the website rarely, change their interests over time, or exhibit different personas. This paper exposes the `Continuous Cold Start' (CoCoS) problem and its consequences for content- and context-based recommendation from the viewpoint of typical e-commerce applications, illustrated with examples from a major travel recommendation website, Booking.com.

cs.IR↗

arXiv2015

Where to Go on Your Next Trip? Optimizing Travel Destinations Based on User Preferences

Recommendation based on user preferences is a common task for e-commerce websites. New recommendation algorithms are often evaluated by offline comparison to baseline algorithms such as recommending random or the most popular items. Here, we investigate how these algorithms themselves perform and compare to the operational production system in large scale online experiments in a real-world application. Specifically, we focus on recommending travel destinations at Booking.com, a major online travel site, to users searching for their preferred vacation activities. To build ranking models we use multi-criteria rating data provided by previous users after their stay at a destination. We implement three methods and compare them to the current baseline in Booking.com: random, most popular, and Naive Bayes. Our general conclusion is that, in an online A/B test with live users, our Naive-Bayes based ranker increased user engagement significantly over the current online system.

cs.IR↗