Many IR systems project harmful societal biases, including gender bias, in their retrieved contents. Uncovering and addressing such biases requires grounded bias measurement principles. However, defining reliable bias metrics for search results is challenging, particularly due to the difficulties in capturing gender-related tendencies in the retrieved documents. In this work, we propose a new framework for search result bias measurement. Within this framework, we first revisit the current metrics for representative search result bias (RepSRB) that are based on the occurrence of gender-specific language in the search results. Addressing their limitations, we additionally propose a metric for comparative search result bias (ComSRB) measurement and integrate it into our framework. ComSRB defines bias as the skew in the set of retrieved documents in response to a non-gendered query toward those for male/female-specific variations of the same query. We evaluate ComSRB against RepSRB on a recent collection of bias-sensitive topics and documents from the MS MARCO collection, using pre-trained bi-encoder and cross-encoder IR models. Our analyses show that, while existing metrics are highly sensitive to the wordings and linguistic formulations, the proposed ComSRB metric mitigates this issue by focusing on the deviations of a retrieval list from its explicitly biased variants, avoiding the need for sub-optimal content analysis processes.

The provided contents by information retrieval (IR) systems can reflect the existing societal biases and stereotypes. Such biases in retrieval results can lead to further establishing and strengthening stereotypes in society and also in the systems. To facilitate the studies of gender bias in the retrieval results of IR systems, we introduce Gender Representation-Bias for Information Retrieval (Grep-BiasIR), a novel thoroughly-audited dataset consisting of 118 bias-sensitive neutral search queries. The set of queries covers a wide range of gender-related topics, for which a biased representation of genders in the search result can be considered as socially problematic. Each query is accompanied with one relevant and one non-relevant document, where the document is also provided in three variations of female, male, and neutral. The dataset is available at https://github.com/KlaraKrieg/GrepBiasIR.

Popularity bias in recommendation lists refers to over-representation of popular content and is a challenge for many recommendation algorithms. Previous research has suggested several offline metrics to quantify popularity bias, which commonly relate the popularity of items in users' recommendation lists to the popularity of items in their interaction history. Discrepancies between these two factors are referred to as popularity miscalibration. While popularity metrics provide a straightforward and well-defined means to measure popularity bias, it is unknown whether they actually reflect users' perception of popularity bias.

We present the LFM-2b dataset containing the listening records of over 120,000 users of the music platform Last.fm. These users provide a total of more than two billion individual listening events that span a time range of over 15 years, from February 2005 until March 2020. These listening events refer to a total of 50 million distinct tracks of 5 million distinct artists. Beside the common metadata (i. e., artist and track name), LFM-2b contains additional information both regarding the users and items. This includes the demographic information of users, namely country, gender, and age, and the fine-grained genre and style of items together with the vector embeddings of their lyrics.

Recent studies show the benefits of reformulating common machine learning models through the concept of prototypes – representatives of the underlying data, used to calculate the prediction score as a linear combination of similarities of a data point to prototypes. Such prototype-based formulation of a model, in addition to preserving (sometimes enhancing) the performance, enables explainability of the model’s decisions, as the prediction can be linearly broken down into the contributions of distinct definable prototypes. Following this direction, we extend the idea of prototypes to the recommender system domain by introducing ProtoMF, a novel collaborative filtering algorithm. ProtoMF learns sets of user/item prototypes that represent the general consumption characteristics of users/items in the underlying dataset. Using these prototypes, ProtoMF then represents users and items as vectors of similarities to the corresponding prototypes. These user/item representations are ultimately leveraged to make recommendations that are both effective in terms of accuracy metrics, and explainable through the interpretation of prototypes’ contributions to the affinity scores. We conduct experiments on three datasets to assess both the effectiveness and the explainability of ProtoMF. Addressing the former, we show that ProtoMF exhibits higher Hit Ratio and NDCG compared to other relevant collaborative filtering approaches. As for the latter, we qualitatively show how ProtoMF can provide explainable recommendations and how its explanation capabilities can expose the existence of statistical biases in the learned representations, which we exemplify for the case of gender bias.

Given a query, neural retrieval models predict point estimates of relevance for each document; however, a significant drawback of relying solely on point estimates is that they contain no indication of the model's confidence in its predictions. Despite this lack of information, downstream methods such as reranking, cutoff prediction, and none-of-the-above classification are still able to learn effective functions to accomplish their respective tasks. Unfortunately, these downstream methods can suffer poor performance when the initial ranking model loses confidence in its score predictions. This becomes increasingly important in high-stakes settings, such as medical searches that can influence health decision making.

Collaborative filtering algorithms capture underlying consumption patterns, including the ones specific to particular demographics or protected information of users, e.g., gender, race, and location. These encoded biases can influence the decision of a recommendation system (RS) towards further separation of the contents provided to various demographic subgroups, and raise privacy concerns regarding the disclosure of users' protected attributes. In this work, we investigate the possibility and challenges of removing specific protected information of users from the learned interaction representations of a RS algorithm, while maintaining its effectiveness. Specifically, we incorporate adversarial training into the state-of-the-art MultVAE architecture, resulting in a novel model, Adversarial Variational Auto-Encoder with Multinomial Likelihood (Adv-MultVAE), which aims at removing the implicit information of protected attributes while preserving recommendation performance. We conduct experiments on the MovieLens-1M and LFM-2b-DemoBias datasets, and evaluate the effectiveness of the bias mitigation method based on the inability of external attackers in revealing the users' gender information from the model. Comparing with baseline MultVAE, the results show that Adv-MultVAE, with marginal deterioration in performance (w.r.t. NDCG and recall), largely mitigates inherent biases in the model on both datasets.

Societal biases can influence Information Retrieval system results, and conversely, search results can potentially reinforce existing societal biases. Recent research has therefore focused on developing methods for quantifying and mitigating bias in search results and applied them to contemporary retrieval systems that leverage transformer-based language models. In the present work, we expand this direction of research by considering bias mitigation within a framework for contextual document embedding reranking. In this framework, the transformer-based query encoder is optimized for relevance ranking through a list-wise objective, by jointly scoring for the same query a large set of candidate document embeddings in the context of one another, instead of in isolation. At the same time, we impose a regularization loss which penalizes highly scoring documents that deviate from neutrality with respect to a protected attribute (e.g., gender). Our approach for bias mitigation is end-to-end differentiable and efficient. Compared to the existing alternatives for deep neural retrieval architectures, which are based on adversarial training, we demonstrate that it can attain much stronger bias mitigation/fairness. At the same time, for the same amount of bias mitigation, it offers significantly better relevance performance (utility). Crucially, our method allows for a more finely controllable and predictable intensity of bias mitigation, which is essential for practical deployment in production systems.

Several studies have identified discrepancies between the popularity of items in user profiles and the corresponding recommendation lists. Such behavior, which concerns a variety of recommendation algorithms, is referred to as popularity bias. Existing work predominantly adopts simple statistical measures, such as the difference of mean or median popularity, to quantify popularity bias. Moreover, it does so irrespective of user characteristics other than the inclination to popular content. In this work, in contrast, we propose to investigate popularity differences (between the user profile and recommendation list) in terms of median, a variety of statistical moments, as well as similarity measures that consider the entire popularity distributions (Kullback-Leibler divergence and Kendall’s τ rank-order correlation). This results in a more detailed picture of the characteristics of popularity bias. Furthermore, we investigate whether such algorithmic popularity bias affects users of different genders in the same way. We focus on music recommendation and conduct experiments on the recently released standardized LFM-2b dataset, containing listening profiles of Last.fm users. We investigate the algorithmic popularity bias of seven common recommendation algorithms (five collaborative filtering and two baselines). Our experiments show that (1) the studied metrics provide novel insights into popularity bias in comparison with only using average differences, (2) algorithms less inclined towards popularity bias amplification do not necessarily perform worse in terms of utility (NDCG), (3) the majority of the investigated recommenders intensify the popularity bias of the female users.

In any ranking system, the retrieval model outputs a single score for a document based on its belief on how relevant it is to a given search query. While retrieval models have continued to improve with the introduction of increasingly complex architectures, few works have investigated a retrieval model's belief in the score beyond the scope of a single value. We argue that capturing the model's uncertainty with respect to its own scoring of a document is a critical aspect of retrieval that allows for greater use of current models across new document distributions, collections, or even improving effectiveness for down-stream tasks. In this paper, we address this problem via an efficient Bayesian framework for retrieval models which captures the model's belief in the relevance score through a stochastic process while adding only negligible computational overhead. We evaluate this belief via a ranking based calibration metric showing that our approximate Bayesian framework significantly improves a retrieval model's ranking effectiveness through a risk aware reranking as well as its confidence calibration. Lastly, we demonstrate that this additional uncertainty information is actionable and reliable on down-stream tasks represented via cutoff prediction.

Societal biases resonate in the retrieved contents of information retrieval (IR) systems, resulting in reinforcing existing stereotypes. Approaching this issue requires established measures of fairness in respect to the representation of various social groups in retrieval results, as well as methods to mitigate such biases, particularly in the light of the advances in deep ranking models. In this work, we first provide a novel framework to measure the fairness in the retrieved text contents of ranking models. Introducing a ranker-agnostic measurement, the framework also enables the disentanglement of the effect on fairness of collection from that of rankers. To mitigate these biases, we propose AdvBert, a ranking model achieved by adapting adversarial bias mitigation for IR, which jointly learns to predict relevance and remove protected attributes. We conduct experiments on two passage retrieval collections (MSMARCO Passage Re-ranking and TREC Deep Learning 2019 Passage Re-ranking), which we extend by fairness annotations of a selected subset of queries regarding gender attributes. Our results on the MSMARCO benchmark show that, (1) all ranking models are less fair in comparison with ranker-agnostic baselines, and (2) the fairness of Bert rankers significantly improves when using the proposed AdvBert models. Lastly, we investigate the trade-off between fairness and utility, showing that we can maintain the significant improvements in fairness without any significant loss in utility.

Click logs are valuable resources for a variety of information retrieval (IR) tasks. This includes query understanding/analysis, as well as learning effective IR models particularly when the models require large amounts of training data. We release a large-scale domain-specific dataset of click logs, obtained from user interactions of the Trip Database health web search engine. Our click log dataset comprises approximately 5.2 million user interactions collected between 2013 and 2020. We use this dataset to create a standard IR evaluation benchmark - TripClick - with around 700,000 unique free-text queries and 1.3 million pairs of query-document relevance signals, whose relevance is estimated by two click-through models. As such, the collection is one of the few datasets offering the necessary data richness and scale to train neural IR models with a large amount of parameters, and notably the first in the health domain. Using TripClick, we conduct experiments to evaluate a variety of IR models, showing the benefits of exploiting this data to train neural architectures. In particular, the evaluation results show that the best performing neural IR model significantly improves the performance by a large margin relative to classical IR models, especially for more frequent queries.

Although recommender systems (RSs) play a crucial role in our society, previous studies have revealed that the performance of RSs may considerably differ between groups of individuals with different characteristics or from different demographics. In this case, a RS is considered to be unfair when it does not perform equally well for different groups of users. Considering the importance of RSs in the distribution and consumption of musical content worldwide, a careful evaluation of fairness in the context of music RSs is crucial. To this end, we first introduce LFM-2b, a novel large-scale real-world dataset of music listening records, comprising a subset to investigate bias of RSs regarding users’ demographics. We then define a notion of fairness based on the performance gap of a RS between the users with different demographics, and evaluate a variety of collaborative filtering algorithms in terms of accuracy and beyond-accuracy metrics to explore the fairness in the RS results toward a specific gender group. We observe the existence of significant discrepancies (unfairness) between the performance of algorithms across male and female user groups. Based on these discrepancies, we explore to what extent recommender algorithms lead to intensifying the underlying population bias in the final results. We also study the effect of a resampling strategy, commonly used as debiasing method , which yields slight improvements in the fairness measures of various algorithms while maintaining their accuracy and beyond-accuracy performance.

The effective extraction of ranked disease-symptom relationships is a critical component in various medical tasks, including computer-assisted medical diagnosis or the discovery of unexpected associations between diseases. While existing disease-symptom relationship extraction methods are used as the foundation in the various medical tasks, no collection is available to systematically evaluate the performance of such methods. In this paper, we introduce the D isease- S ymptom R elation Collection ( dsr -collection), created by five physicians as expert annotators. We provide graded symptom judgments for diseases by differentiating between relevant symptoms and primary symptoms . Further, we provide several strong baselines, based on the methods used in previous studies. The first method is based on word embeddings, and the second on co-occurrences of MeSH-keywords of medical articles. For the co-occurrence method, we propose an adaption in which not only keywords are considered, but also the full text of medical articles. The evaluation on the dsr -collection shows the effectiveness of the proposed adaption in terms of nDCG, precision, and recall.

Concerns regarding the footprint of societal biases in information retrieval (IR) systems have been raised in several previous studies. In this work, we examine various recent IR models from the perspective of the degree of gender bias in their retrieval results. To this end, we first provide a bias measurement framework which includes two metrics to quantify the degree of the unbalanced presence of gender-related concepts in a given IR model's ranking list. To examine IR models by means of the framework, we create a dataset of non-gendered queries, selected by human annotators. Applying these queries to the MS MARCO Passage retrieval collection, we then measure the gender bias of a BM25 model and several recent neural ranking models. The results show that while all models are strongly biased toward male, the neural models, and in particular the ones based on contextualized embedding models, significantly intensify gender bias. Our experiments also show an overall increase in the gender bias of neural models when they exploit transfer learning, namely when they use (already biased) pre-trained embeddings.

Low-frequency terms are a recurring challenge for information retrieval models, especially neural IR frameworks struggle with adequately capturing infrequently observed words. While these terms are often removed from neural models - mainly as a concession to efficiency demands - they traditionally play an important role in the performance of IR models. In this paper, we analyze the effects of low-frequency terms on the performance and robustness of neural IR models. We conduct controlled experiments on three recent neural IR models, trained on a large-scale passage retrieval collection. We evaluate the neural IR models with various vocabulary sizes for their respective word embeddings, considering different levels of constraints on the available GPU memory. We observe that despite the significant benefits of using larger vocabularies, the performance gap between the vocabularies can be, to a great extent, mitigated by extensive tuning of a related parameter: the number of documents to re-rank. We further investigate the use of subword-token embedding models, and in particular FastText, for neural IR models. Our experiments show that using FastText brings slight improvements to the overall performance of the neural IR models in comparison to models trained on the full vocabulary, while the improvement becomes much more pronounced for queries containing low-frequency terms.

Exploitation of term relatedness provided by word embedding has gained considerable attention in recent IR literature. However, an emerging question is whether this sort of relatedness fits to the needs of IR with respect to retrieval effectiveness. While we observe a high potential of word embedding as a resource for related terms, the incidence of several cases of topic shifting deteriorates the final performance of the applied retrieval models. To address this issue, we revisit the use of global context (i.e. the term co-occurrence in documents) to measure the term relatedness. We hypothesize that in order to avoid topic shifting among the terms with high word embedding similarity, they should often share similar global contexts as well. We therefore study the effectiveness of post filtering of related terms by various global context relatedness measures. Experimental results show significant improvements in two out of three test collections, and support our initial hypothesis regarding the importance of considering global context in retrieval.

A recurring question in information retrieval is whether term associations can be properly integrated in traditional information retrieval models while preserving their robustness and effectiveness. In this paper, we revisit a wide spectrum of existing models (Pivoted Document Normalization, BM25, BM25 Verboseness Aware, Multi-Aspect TF, and Language Modelling) by introducing a generalisation of the idea of the translation model. This generalisation is a de facto transformation of the translation models from Language Modelling to the probabilistic models. In doing so, we observe a potential limitation of these generalised translation models: they only affect the term frequency based components of all the models, ignoring changes in document and collection statistics. We correct this limitation by extending the translation models with the 15 statistics of term associations and provide extensive experimental results to demonstrate the benefit of the newly proposed methods. Additionally, we compare the translation models with query expansion methods based on the same term association resources, as well as based on Pseudo-Relevance Feedback (PRF). We observe that translation models always outperform the first, but provide complementary information with the second, such that by using PRF and our translation models together we observe results better than the current state of the art.

Recent developments on word embedding provide a novel source of information for term-to-term similarity. A recurring question now is whether the provided term associations can be properly integrated in the traditional information retrieval models while preserving their robustness and effectiveness. In this paper, we propose addressing the question of combining the term-to-term similarity of word embedding with IR models. The retrieval models in the approach are enhanced by altering the basic components of document retrieval, i.e. term frequency (tf) and document frequency (df). In addition, we target the study of the meaning of the term relatedness of word embedding models and its applicability in IR. This research topic consists of first explore of reliable similarity thresholds of word embedding vectors to indicate ?related terms? and second, identification of the linguistic types of the terms relatedness.