Classification models in machine learning are typically trained using coarse-grained class labels. Although these models often achieve strong overall accuracy, their performance is asymmetric across the subclasses that arise out of a common phenomenon called hidden stratification. Generally, the latent subclasses within each class differ substantially in distribution and characteristics, resulting in poor generalization for underrepresented groups. Moreover, imbalanced subclass distributions lead to majority subclasses dominating training, resulting in biased and less reliable models, especially for safety-critical applications (such as medical). To address these challenges, we propose a novel framework that attempts to uncover hidden subclasses via a repulsive point process. Our approach then leverages these fine-grained labels to make the classifier more inclusive across the subclasses. Our approach identifies subclasses without requiring additional supervision, thereby promoting diversity and reducing sensitivity to subclass imbalance. Extensive experiments on four benchmark datasets demonstrate consistent and significant improvements over state-of-the-art baselines across both balanced and imbalanced subclass distributions, underscoring the effectiveness and generalizability of our approach.

In recent years, with the emergence of online learning platforms and e-learning resources, many documents are available for a particular topic. For a better learning experience, the learner often needs to know and learn first the prerequisite concepts for a given concept. Traditionally, the identification of such prerequisite concepts is done manually by subject experts, which in turn, often limits self-paced learning. Recently, machine learning models have found encouraging success for the task, obviating manual effort. In this paper, we propose a graph neural network based approach that leverages node attention over a heterogeneous graph to extract the prerequisite concepts for a given concept. Experiments on a set of benchmark data show that the proposed model outperforms the existing models by large margins almost always, making the model a new state-of-the-art for the task.

Abstract Inferring the gender of named entities present in a text has several practical applications in information sciences. Existing approaches toward name gender identification rely exclusively on using the gender distributions from labeled data. In the absence of such labeled data, these methods fail. In this article, we propose a two‐stage model that is able to infer the gender of names present in text without requiring explicit name‐gender labels. We use coreference resolution as the backbone for our proposed model. To aid coreference resolution where the existing contextual information does not suffice, we use a retrieval‐assisted context aggregation framework. We demonstrate that state‐of‐the‐art name gender inference is possible without supervision. Our proposed method matches or outperforms several supervised approaches and commercially used methods on five English language datasets from different domains.

Existing probabilistic retrieval models do not restrict the domain of the random variables that they deal with. In this article, we show that the upper bound of the normalized term frequency ( tf ) from the relevant documents is much smaller than the upper bound of the normalized tf from the whole collection. As a result, the existing models suffer from two major problems: (i) the domain mismatch causes data modeling error, (ii) since the outliers have very large magnitude and the retrieval models follow tf hypothesis, the combination of these two factors tends to overestimate the relevance score. In an attempt to address these problems, we propose novel weighted probabilistic models based on truncated distributions. We evaluate our models on a set of large document collections. Significant performance improvement over six existing probabilistic models is demonstrated.

This article proposes a term weighting scheme for measuring query-document similarity that attempts to explicitly model the dependency between separate occurrences of a term in a document. The assumption is that, if a term appears once in a document, it is more likely to appear again in the same document. Thus, as the term appears again and again, the information content of the subsequent occurrences decreases gradually, since they are more predictable. We introduce a parameterized decay function to model this assumption, where the initial contribution of the term can be determined using any reasonable term discrimination factor. The effectiveness of the proposed model is evaluated on a number of recent web test collections of varying nature. The experimental results show that the proposed model significantly outperforms a number of well known retrieval models including a recently proposed strong Term Frequency and Inverse Document Frequency (TF-IDF) model.

The main goal of a retrieval model is to measure the degree of relevance of a document with respect to the given query. Probabilistic models are widely used to measure the likelihood of relevance of a document by combining within document term frequency and term specificity in a formal way. Recent research shows that tf normalization that factors in multiple aspects of term salience is an effective scheme. However, existing models do not fully utilize these tf normalization components in a principled way. Moreover, most state of the art models ignore the distribution of a term in the part of the collection that contains the term. In this article, we introduce a new probabilistic model of ranking that addresses the above issues. We argue that, since the relevance of a document increases with the frequency of the query term, this assumption can be used to measure the likelihood that the normalized frequency of a term in a particular document will be maximum with respect to its distribution in the elite set. Thus, the weight of a term in a document is proportional to the probability that the normalized frequency of that term is maximum under the hypothesis that the frequencies are generated randomly. To that end, we introduce a ranking function based on maximum value distribution that uses two aspects of tf normalization. The merit of the proposed model is demonstrated on a number of recent large web collections. Results show that the proposed model outperforms the state of the art models by significantly large margin.

The multinomial language model has been one of the most effective models of retrieval for more than a decade. However, the multinomial distribution does not model one important linguistic phenomenon relating to term dependency—that is, the tendency of a term to repeat itself within a document (i.e., word burstiness). In this article, we model document generation as a random process with reinforcement (a multivariate Pólya process) and develop a Dirichlet compound multinomial language model that captures word burstiness directly. We show that the new reinforced language model can be computed as efficiently as current retrieval models, and with experiments on an extensive set of TREC collections, we show that it significantly outperforms the state-of-the-art language model for a number of standard effectiveness metrics. Experiments also show that the tuning parameter in the proposed model is more robust than that in the multinomial language model. Furthermore, we develop a constraint for the verbosity hypothesis and show that the proposed model adheres to the constraint. Finally, we show that the new language model essentially introduces a measure closely related to idf, which gives theoretical justification for combining the term and document event spaces in tf-idf type schemes.

The term weighting and document ranking functions used with informational queries are typically optimized for cases in which queries are short and documents are long. It is reasonable to assume that the presence of a term in a short query reflects some aspect of the topic that is important to the user, and thus rewarding documents that contain the greatest number of distinct query terms is a useful heuristic. Verbose informational queries, such as those that result from cut-and-paste of example text, or that might result from informal spoken interaction, pose a different challenge in which many extraneous (and thus potentially misleading) terms may be present in the query. Modest improvements have been reported from applying supervised methods to learn which terms in a verbose query deserve the greatest emphasis. This paper proposes a novel unsupervised method for weighting terms in verbose informational queries that relies instead on iteratively estimating which terms are most central to the query. The key idea is to use an initial set of retrieval results to define a recursion on the term weight vector that converges to a fixed point representing the vector that optimally describes the initial result set. Experiments with several TREC news and Web test collections indicate that the proposed method often statistically significantly outperforms state of the art supervised methods.

In some jurisdictions, parties to a lawsuit can request documents from each other, but documents subject to a claim of privilege may be withheld. The TREC 2010 Legal Track developed what is presently the only public test collection for evaluating privilege classification. This paper examines the reliability and reusability of that collection. For reliability, the key question is the extent to which privilege judgments correctly reflect the opinion of the senior litigator whose judgment is authoritative. For reusability, the key question is the degree to which systems whose results contributed to creation of the test collection can be fairly compared with other systems that use those privilege judgments in the future. These correspond to measurement error and sampling error, respectively. The results indicate that measurement error is the larger problem.

Term weighting schemes are central to the study of information retrieval systems. This article proposes a novel TF-IDF term weighting scheme that employs two different within document term frequency normalizations to capture two different aspects of term saliency. One component of the term frequency is effective for short queries, while the other performs better on long queries. The final weight is then measured by taking a weighted combination of these components, which is determined on the basis of the length of the corresponding query.

Stemming is a widely used technique in information retrieval systems to address the vocabulary mismatch problem arising out of morphological phenomena. The major shortcoming of the commonly used stemmers is that they accept the morphological variants of the query words without considering their thematic coherence with the given query, which leads to poor performance. Moreover, for many queries, such approaches also produce retrieval performance that is poorer than no stemming, thereby degrading the robustness. The main goal of this article is to present corpus-based fully automatic stemming algorithms which address these issues. A set of experiments on six TREC collections and three other non-English collections containing news and web documents shows that the proposed query-based stemming algorithms consistently and significantly outperform four state of the art strong stemmers of completely varying principles. Our experiments also confirm that the robustness of the proposed query-based stemming algorithms are remarkably better than the existing strong baselines.

We present a stemming algorithm for text retrieval. The algorithm uses the statistics collected on the basis of certain corpus analysis based on the co-occurrence between two word variants. We use a very simple co-occurrence measure that reflects how often a pair of word variants occurs in a document as well as in the whole corpus. A graph is formed where the word variants are the nodes and two word variants form an edge if they co-occur. On the basis of the co-occurrence measure, a certain edge strength is defined for each of the edges. Finally, on the basis of the edge strengths, we propose a partition algorithm that groups the word variants based on their strongest neighbors, that is, the neighbors with largest strengths.

A novel graph-based language-independent stemming algorithm suitable for information retrieval is proposed in this article. The main features of the algorithm are retrieval effectiveness, generality, and computational efficiency. We test our approach on seven languages (using collections from the TREC, CLEF, and FIRE evaluation platforms) of varying morphological complexity. Significant performance improvement over plain word-based retrieval, three other language-independent morphological normalizers, as well as rule-based stemmers is demonstrated.