The problem of fair decision had been studied within information systems for a long time. In the event of Information Retrieval and Recommendation, the need of accessing information by means of rankings complicates the search for fairness because the exposure of an item depends on the user's attention and on the rank of the item. Starting from the observation that a ranking is the result of a series of preferences based on a variety of criteria, this paper describes Preference Fair Ranking (PFR). In a nutshell, matrices are defined containing the preferences between items according to different criteria and the eigenvectors of a linear combination of the matrices that establish the best ranking are found. The paper also reports on the experiments carried out by using two large public test collections for fair information retrieval, i.e. 2021 and 2022 TREC Fair tracks and by comparing the proposed method with two well-known fair ranking algorithms. The results show that the proposed method significantly outperform the other two methods. In particular, PFR showed to be superior in fairness to the baseline methods in terms of the TREC Fair track measures. Besides retrieval effectiveness and fairness, PFR is computationally efficient and helps the flexible design of multiple criteria ranking.

Abstract The requirement that, for fair document retrieval, the documents should be ranked in the order to equally expose authors and organizations has been studied for some years. The fair exposure of a ranking, however, undermines the optimality of the Probability Ranking Principle and as a consequence retrieval effectiveness. It is shown how the variations of fairness and effectiveness can be related by a model. To this end, the paper introduces a fairness measure inspired in Gini’s index of mutability for non-ordinal variables and relates it to a general enough measure of effectiveness, thus modeling the connection between these two dimensions of Information Retrieval. The paper also introduces the measurement of the statistical significance of the fairness measure. An empirical study completes the paper.

Video sentiment analysis as a decision-making process is inherently complex, involving the fusion of decisions from multiple modalities and the so-caused cognitive biases. Inspired by recent advances in quantum cognition, we show that the sentiment judgment from one modality could be incompatible with the judgment from another, i.e., the order matters and they cannot be jointly measured to produce a final decision. Thus the cognitive process exhibits ``quantum-like'' biases that cannot be captured by classical probability theories. Accordingly, we propose a fundamentally new, quantum cognitively motivated fusion strategy for predicting sentiment judgments. In particular, we formulate utterances as quantum superposition states of positive and negative sentiment judgments, and uni-modal classifiers as mutually incompatible observables, on a complex-valued Hilbert space with positive-operator valued measures. Experiments on two benchmarking datasets illustrate that our model significantly outperforms various existing decision level and a range of state-of-the-art content-level fusion approaches. The results also show that the concept of incompatibility allows effective handling of all combination patterns, including those extreme cases that are wrongly predicted by all uni-modal classifiers.

We provide a novel perspective on conversational emotion recognition by drawing an analogy between the task and a complete span of quantum measurement. We characterize different steps of quantum measurement in the process of recognizing speakers' emotions in conversation, and stitch them up with a quantum-like neural network. The quantum-like layers are implemented by complex-valued operations to ensure an authentic adoption of quantum concepts, which naturally enables conversational context modeling and multimodal fusion. We borrow an existing algorithm to learn the complex-valued network weights, so that the quantum-like procedure is conducted in a data-driven manner. Our model is comparable to state-of-the-art approaches on two benchmarking datasets, and provide a quantum view to understand conversational emotion recognition.

No abstract available.

Word meaning may change over time as a reflection of changes in human society. Therefore, modeling time in word representation is necessary for some diachronic tasks. Most existing diachronic word representation approaches train the embeddings separately for each pre-grouped time-stamped corpus and align these embeddings, e.g., by orthogonal projections, vector initialization, temporal referencing, and compass. However, not only does word meaning change in a short time, word meaning may also be subject to evolution over long timespans, thus resulting in a unified continuous process. A recent approach called `DiffTime' models semantic evolution as functions parameterized by multiple-layer nonlinear neural networks over time. In this paper, we will carry on this line of work by learning explicit functions over time for each word. Our approach, called `Word2Fun', reduces the space complexity from $\mathcal{O}(TVD)$ to $\mathcal{O}(kVD)$ where $k$ is a small constant ($k \ll T $). In particular, a specific instance based on polynomial functions could provably approximate any function modeling word evolution with a given negligible error thanks to the Weierstrass Approximation Theorem. The effectiveness of the proposed approach is evaluated in diverse tasks including time-aware word clustering, temporal analogy, and semantic change detection. Code at: {\url{https://github.com/wabyking/Word2Fun.git}}.

The BIRDS workshop aimed to foster the cross-fertilization of Information Science (IS), Information Retrieval (IR) and Data Science (DS). Recognising the commonalities and differences between these communities, the proposed full-day workshop brought together experts and researchers in IS, IR and DS to discuss how they can learn from each other to provide more user-driven data and infor- mation exploration and retrieval solutions. Therefore, the papers aimed to convey ideas on how to utilise, for instance, IS concepts and theories in DS and IR or DS approaches to support users in data and information exploration.

Machine Learning (ML) helps us to recognize patterns from raw data. ML is used in numerous domains i.e. biomedical, agricultural, food technology, etc. Despite recent technological advancements, there is still room for substantial improvement in prediction. Current ML models are based on classical theories of probability and statistics, which can now be replaced by Quantum Theory (QT) with the aim of improving the effectiveness of ML. In this paper, we propose the Binary Classifier Inspired by Quantum Theory (BCIQT) model, which outperforms the state of the art classification in terms of recall for every category.

Capturing the meaning of sentences has long been a challenging task. Current models tend to apply linear combinations of word features to conduct semantic composition for bigger-granularity units e.g. phrases, sentences, and documents. However, the semantic linearity does not always hold in human language. For instance, the meaning of the phrase “ivory tower” cannot be deduced by linearly combining the meanings of “ivory” and “tower”. To address this issue, we propose a new framework that models different levels of semantic units (e.g. sememe, word, sentence, and semantic abstraction) on a single Semantic Hilbert Space, which naturally admits a non-linear semantic composition by means of a complex-valued vector word representation. An end-to-end neural network 1 is proposed to implement the framework in the text classification task, and evaluation results on six benchmarking text classification datasets demonstrate the effectiveness, robustness and self-explanation power of the proposed model. Furthermore, intuitive case studies are conducted to help end users to understand how the framework works.

Learning algorithms based on probability organize the observed data in subsets corresponding to binary variables. In this paper, we address the problem of estimating one probability space given a set of observed data about n variables or properties. One problem with estimating one single probability space is the exponential number of events. Approximation is one approach to addressing the problem of the exponential order of the number of events. Alternatively to approximation, we change paradigm - from classical, set-based probability spaces based on sets to quantum probability spaces based on vector subspaces. By changing paradigm, we leverage quantum probability and present an efficient algorithm to calculate a Quantum Probability Space (QPS) in only O(n4) steps.

Machine Learning models learn the relationship between input and output by examples and then apply the learned models to relate unseen input. Although ML has successfully been used in almost every field, there is always room for improvement. To this end, researchers have recently been trying to implement Quantum Mechanics(QM) in ML, since it is believed that quantum-inspired ML can enhance learning rate and effectiveness. In this paper, we address a specific task of ML and present a binary classification model inspired by the quantum detection framework. We compared the model to the state of the art. Our experimental results suggest that the use of the quantum detection framework in binary classification can improve effectiveness for a number of topics of the RCV-1 test collection and that it may still provide ways to improve effectiveness for the other topics.

Information Retrieval (IR) is the complex of activities that represent information as data and rank the data representing information relevant to the user's information needs by a retrieval function. Such a function involves parameters. They can in principle be set irrespective of the specific set of documents and queries, but can in practice maximize retrieval effectiveness. However, algorithms to select retrieval function parameters must be efficient due to the large search space. We can remark that: (i) all the tested methods are similarly effective, but the plots of the maximum value of NDCG@20 at a given evaluation show that our algorithm is more efficient; (ii) performance metrics and datasets studied in this paper seem to yield objective functions with few, if any, local optima with large basin of attraction; (iii) our algorithm is considerably more efficient, quickly finding parameterizations of the retrieval function yielding high performance - much faster than line search.

The retrieval function is one of the most important components of an Information Retrieval (IR) system, because it determines to what extent some information is relevant to a user query. Most retrieval functions have “free parameters” whose value must be set before retrieval, significantly affecting the effectiveness of an IR system. Choosing the optimum values for such parameters is therefore of paramount importance. However, the optimum can only be found after a computationally expensive process, especially when the generalization error is estimated via cross-validation. In this paper, we propose to determine free parameter values by solving an optimization problem aimed at maximizing a measure of retrieval effectiveness. We employ the black-box optimization paradigm, since the analytical expression of the measure of effectiveness with respect to the free parameters is unknown. We consider different methods for solving the black-box optimization problem: a simple grid-search over the whole domain, and more sophisticated techniques such as line search and surrogate model based algorithms. Experimental results on several test collections not only provide useful insight about effectiveness, but also about efficiency: they indicate that with appropriate optimization techniques, the computational cost of parameter optimization can be greatly reduced without compromising retrieval effectiveness, even when taking generalization into account.

Information Retrieval (IR) effectiveness measures commonly assume that the experimental query sets consist of randomly drawn queries that represent the population of queries submitted to IR systems. In many practical situations, however, this assumption is violated, in a problem known as sample selection bias. It follows that the systems participating in evaluation campaigns are ranked by biased estimators of effectiveness. In this paper, we address the problem of query sample selection bias in machine learning terms and study experimentally how retrieval system rankings are affected by it. To this end, we apply a number of retrieval effectiveness measures and query probability estimation methods useful to correct sample selection bias. We report that the ranking of the most effective systems and that of the least effective systems is fairly affected by query sample selection bias, while the ranking of the average systems is much more affected. We also report that the measure of bias depends on the retrieval measure used to rank systems and eventually on the search task being evaluated.

Information Retrieval (IR) is concerned with indexing and retrieving documents including information relevant to a user's information need. Relevance Feedback (RF) is a class of effective algorithms for improving Information Retrieval (IR) and it consists of gathering further data representing the user's information need and automatically creating a new query. In this paper, we propose a class of RF algorithms inspired by quantum detection to re-weight the query terms and to re-rank the document retrieved by an IR system. These algorithms project the query vector on a subspace spanned by the eigenvector which maximizes the distance between the distribution of quantum probability of relevance and the distribution of quantum probability of non-relevance. The experiments showed that the RF algorithms inspired by quantum detection can outperform the state-of-the-art algorithms.

Information retrieval (IR) systems are designed, in general, to satisfy the information need of a user who expresses it by means of a query, by providing him with a subset of documents selected from a collection and ordered by decreasing relevance to the query. Such systems are based on IR models, which define how to represent the documents and the query, as well as how to determine the relevance of a document for a query. In this article, we present a new IR model based on concepts taken from both IR and digital signal processing (like Fourier analysis of signals and filtering). This allows the whole IR process to be seen as a physical phenomenon, where the query corresponds to a signal, the documents correspond to filters, and the determination of the relevant documents to the query is done by filtering that signal. Tests showed that the quality of the results provided by this IR model is comparable with the state-of-the-art.

Information Retrieval (IR) is concerned with indexing and retrieving documents including information relevant to a user's information need. Relevance Feedback (RF) is an effective technique for improving IR and it consists of gathering further data representing the user's information need and automatically creating a new query. As RF relies on the ability of an IR system to learn new queries and is mostly based on statistical methods, a parallel between RF and statistical Machine Learning (ML) can be drawn. However, the effectiveness of RF is due to the biased selection of the sample data, thus contradicting the requirement that effective statistical learning is based on unbiased sample data. This paper studies this contradiction and suggests that RF cannot be straightforwardly studied within statistical ML without considering the intrinsic nature of the data managed by an IR system and of the user's information need. In particular, the paper reports that an RF algorithm is mostly influenced by the distance between the informative content of the training set and the informative content of the test set and is not influenced by sample selection bias.

Indexing is a core process of an information retrieval (IR) system (IRS). As indexing can neither be exhaustive nor precise, the decision taken by an IRS about the relevance of the content of a document to an information need is subject to uncertainty. It is our hypothesis that one of the reasons that IRSs are unable to optimally respond to every query is that the document collections and the posting lists are modeled as sets of documents. In contrast, if vector spaces replace sets along the way given by quantum mechanics, it is possible to define a quantum information retrieval basis (QIRB) that at least in principle yields more effective document ranking than the ranking yielded by the current principles, with effectiveness being measured in terms of recall at every level of fallout. To this end, we show that the probability ranking principle and the Neyman–Pearson Lemma (NPL) are equivalent. The rest of the article follows from this result. In particular, we introduce the QIRB, link it to a generalization of the NPL and demonstrate its superiority through a concise mathematical analysis and an empirical study. The challenges posed by this basis and the research directions that would be opened are also discussed.

Information retrieval (IR) models based on vector spaces have been investigated for a long time. Nevertheless, they have recently attracted much research interest. In parallel, context has been rediscovered as a crucial issue in information retrieval. This article presents a principled approach to modeling context and its role in ranking information objects using vector spaces. First, the article outlines how a basis of a vector space naturally represents context, both its properties and factors. Second, a ranking function computes the probability of context in the objects represented in a vector space, namely, the probability that a contextual factor has affected the preparation of an object.

Implicit feedback algorithms utilize interaction between searchers and search systems to learn more about users' needs and interests than expressed in query statements alone. This additional information can be used to formulate improved queries or directly improve retrieval performance. In this paper we present a geometric framework that utilizes multiple sources of evidence present in this interaction context (e.g., display time, document retention) to develop enhanced implicit feedback models personalized for each user and tailored for each search task. We use rich interaction logs (and associated metadata such as relevance judgments), gathered during a longitudinal user study, as relevance stimuli to compare an implicit feedback algorithm developed using the framework with alternative algorithms. Our findings demonstrate both the effectiveness of our proposed algorithm and the potential value of incorporating multiple sources of interaction evidence when developing implicit feedback algorithms.

Abstract The authors describe a statistical approach based on hidden Markov models (HMMs), for generating stemmers automatically. The proposed approach requires little effort to insert new languages in the system even if minimal linguistic knowledge is available. This is a key advantage especially for digital libraries, which are often developed for a specific institution or government because the program can manage a great amount of documents written in local languages. The evaluation described in the article shows that the stemmers implemented by means of HMMs are as effective as those based on linguistic rules.

No abstract available.

In this paper, context is modeled by vector space bases and its evolution is modeled by linear transformations from one base to another. Each document or query can be associated to a distinct base, which corresponds to one context. Also, algorithms are proposed to discover contexts from document, query or groups or them. Linear algebra can thus by employed in a mathematical framework to process context, its evolution and application.

Abstract Textbooks are more available in electronic format now than in the past. Because textbooks are typically large, the end user needs effective tools to rapidly access information encapsulated in textbooks stored in digital libraries. Statistical similarity‐based links among hyper‐textbooks are a means to provide those tools. In this paper, the design and the implementation of a tool that generates networks of links within and across hyper‐textbooks through a completely automatic and unsupervised procedure is described. The design is based on statistical techniques. The overall methodology is presented together with the results of a case study reached through a working prototype that shows that connecting hyper‐textbooks is an efficient way to provide an effective retrieval capability.

Abstract The article describes the project on music information retrieval that has been carried out at the University of Padova, Italy. The research work has been characterized by the synergy of the modular integration of sound techniques of melody processing and of statistical information retrieval. After illustrating the background from which the project has originated, we describe the complete process, from methodology design through evaluation and system implementation. Conclusions, impacts on research in music information retrieval, and future directions are also described.

In this paper, we present a method based on Hidden Markov Models (HMMs) to generate statistical stemmers. Using a list of words as training set, the method estimates the HMM parameters which are used to calculate the most probable stem for an arbitrary word. Stemming is performed by computing the most probable path, through the HMM states, corresponding to the input word. Linguistic knowledge or a training set of manually stemmed words are not required. We describe the method and the results of the experiments carried out using standard test collections for five different languages.

No abstract available.