Fairness of exposure is a commonly used notion of fairness for ranking systems. It is based on the idea that all items or item groups should get exposure proportional to the merit of the item or the collective merit of the items in the group. Often, stochastic ranking policies are used to ensure fairness of exposure. Previous work unrealistically assumes that we can reliably estimate the expected exposure for all items in each ranking produced by the stochastic policy. In this work, we discuss how to approach fairness of exposure in cases where the policy contains rankings of which, due to inter-item dependencies, we cannot reliably estimate the exposure distribution. In such cases, we cannot determine whether the policy can be considered fair. Our contributions in this paper are twofold. First, we define a method called FELIX for finding stochastic policies that avoid showing rankings with unknown exposure distribution to the user without having to compromise user utility or item fairness. Second, we extend the study of fairness of exposure to the top-k setting and also assess FELIX in this setting. We find that FELIX can significantly reduce the number of rankings with unknown exposure distribution without a drop in user utility or fairness compared to existing fair ranking methods, both for full-length and top-k rankings. This is an important first step in developing fair ranking methods for cases where we have incomplete knowledge about the user's behaviour.
The cross-entropy objective has proved to be an all-purpose training objective for autoregressive language models (LMs). However, without considering the penalization of problematic tokens, LMs trained using cross-entropy exhibit text degeneration. To address this, unlikelihood training has been proposed to reduce the probability of unlikely tokens predicted by LMs. But unlikelihood does not consider the relationship between the label tokens and unlikely token candidates, thus showing marginal improvements in degeneration. We propose a new contrastive token learning objective that inherits the advantages of cross-entropy and unlikelihood training and avoids their limitations. The key idea is to teach a LM to generate high probabilities for label tokens and low probabilities of negative candidates. Comprehensive experiments on language modeling and open-domain dialogue generation tasks show that the proposed contrastive token objective yields much less repetitive texts, with a higher generation quality than baseline approaches, achieving the new state-of-the-art performance on text degeneration.
Methods for reinforcement learning for recommendation (RL4Rec) are increasingly receiving attention as they can quickly adapt to user feedback. A typical RL4Rec framework consists of (1) a state encoder to encode the state that stores the users' historical interactions, and (2) an RL method to take actions and observe rewards. Prior work compared four state encoders in an environment where user feedback is simulated based on real-world logged user data. An attention-based state encoder was found to be the optimal choice as it reached the highest performance. However, this finding is limited to the actor-critic method, four state encoders, and evaluation-simulators that do not debias logged user data. In response to these shortcomings, we reproduce and expand on the existing comparison of attention-based state encoders (1) in the publicly available debiased RL4Rec SOFA simulator with (2) a different RL method, (3) more state encoders, and (4) a different dataset. Importantly, our experimental results indicate that existing findings do not generalize to the debiased SOFA simulator generated from a different dataset and a Deep Q-Network (DQN)-based method when compared with more state encoders.
There are several measures for fairness in ranking, based on different underlying assumptions and perspectives. PL optimization with the REINFORCE algorithm can be used for optimizing black-box objective functions over permutations. In particular, it can be used for optimizing fairness measures. However, though effective for queries with a moderate number of repeating sessions, PL optimization has room for improvement for queries with a small number of repeating sessions. In this paper, we present a novel way of representing permutation distributions, based on the notion of permutation graphs. Similar to PL, our distribution representation, called PPG, can be used for black-box optimization of fairness. Different from PL, where pointwise logits are used as the distribution parameters, in PPG pairwise inversion probabilities together with a reference permutation construct the distribution. As such, the reference permutation can be set to the best sampled permutation regarding the objective function, making PPG suitable for both deterministic and stochastic rankings. Our experiments show that PPG, while comparable to PL for larger session repetitions (i.e., stochastic ranking), improves over PL for optimizing fairness metrics for queries with one session (i.e., deterministic ranking). Additionally, when accurate utility estimations are available, e.g., in tabular models, the performance of PPG in fairness optimization is significantly boosted compared to lower quality utility estimations from a learning to rank model, leading to a large performance gap with PL. Finally, the pairwise probabilities make it possible to impose pairwise constraints such as "item $d_1$ should always be ranked higher than item $d_2$." Such constraints can be used to simultaneously optimize the fairness metric and control another objective such as ranking performance.
To train image-caption retrieval (ICR) methods, contrastive loss functions are a common choice for optimization functions. Unfortunately, contrastive ICR methods are vulnerable to learning shortcuts: decision rules that perform well on the training data but fail to transfer to other testing conditions. We introduce an approach to reduce shortcut feature representations for the ICR task: latent target decoding (LTD). We add an additional decoder to the learning framework to reconstruct the input caption, which prevents the image and caption encoder from learning shortcut features. Instead of reconstructing input captions in the input space, we decode the semantics of the caption in a latent space. We implement the LTD objective as an optimization constraint, to ensure that the reconstruction loss is below a threshold value while primarily optimizing for the contrastive loss. Importantly, LTD does not depend on additional training data or expensive (hard) negative mining strategies. Our experiments show that, unlike reconstructing the input caption, LTD reduces shortcut learning and improves generalizability by obtaining higher recall@k and r-precision scores. Additionally, we show that the evaluation scores benefit from implementing LTD as an optimization constraint instead of a dual loss.
$ $Dialogue systems are evaluated depending on their type and purpose. Two categories are often distinguished: (1) task-oriented dialogue systems (TDS), which are typically evaluated on utility, i.e., their ability to complete a specified task, and (2) open domain chatbots, which are evaluated on the user experience, i.e., based on their ability to engage a person. What is the influence of user experience on the user satisfaction rating of TDS as opposed to, or in addition to, utility? We collect data by providing an additional annotation layer for dialogues sampled from the ReDial dataset, a widely used conversational recommendation dataset. Unlike prior work, we annotate the sampled dialogues at both the turn and dialogue level on six dialogue aspects: relevance, interestingness, understanding, task completion, efficiency, and interest arousal. The annotations allow us to study how different dialogue aspects influence user satisfaction. We introduce a comprehensive set of user experience aspects derived from the annotators' open comments that can influence users' overall impression. We find that the concept of satisfaction varies across annotators and dialogues, and show that a relevant turn is significant for some annotators, while for others, an interesting turn is all they need. Our analysis indicates that the proposed user experience aspects provide a fine-grained analysis of user satisfaction that is not captured by a monolithic overall human rating.
Automatic text summarization has experienced substantial progress in recent years. With this progress, the question has arisen whether the types of summaries that are typically generated by automatic summarization models align with users' needs. Ter Hoeve et al (2020) answer this question negatively. Amongst others, they recommend focusing on generating summaries with more graphical elements. This is in line with what we know from the psycholinguistics literature about how humans process text. Motivated from these two angles, we propose a new task: summarization with graphical elements, and we verify that these summaries are helpful for a critical mass of people. We collect a high quality human labeled dataset to support research into the task. We present a number of baseline methods that show that the task is interesting and challenging. Hence, with this work we hope to inspire a new line of research within the automatic summarization community.
A long-term ambition of information seeking QA systems is to reason over multi-modal contexts and generate natural answers to user queries. Today, memory intensive pre-trained language models are adapted to downstream tasks such as QA by fine-tuning the model on QA data in a specific modality like unstructured text or structured tables. To avoid training such memory-hungry models while utilizing a uniform architecture for each modality, parameter-efficient adapters add and train small task-specific bottle-neck layers between transformer layers. In this work, we study parameter-efficient abstractive QA in encoder-decoder models over structured tabular data and unstructured textual data using only 1.5% additional parameters for each modality. We also ablate over adapter layers in both encoder and decoder modules to study the efficiency-performance trade-off and demonstrate that reducing additional trainable parameters down to 0.7%-1.0% leads to comparable results. Our models out-perform current state-of-the-art models on tabular QA datasets such as Tablesum and FeTaQA, and achieve comparable performance on a textual QA dataset such as NarrativeQA using significantly less trainable parameters than fine-tuning.