We explore the feasibility of automatically identifying sentences in different MEDLINE abstracts that are related in meaning. We compared traditional vector space models with machine learning methods for detecting relatedness, and found that machine learning was superior. The Huber method, a variant of Support Vector Machines which minimizes the modified Huber loss function, achieves 73% precision when the score cutoff is set high enough to identify about one related sentence per abstract on average. We illustrate how an abstract viewed in PubMed might be modified to present the related sentences found in other abstracts by this automatic procedure.

For the purposes of classification it is common to represent a document as a bag of words. Such a representation consists of the individual terms making up the document together with the number of times each term appears in the document. All classification methods make use of the terms. It is common to also make use of the local term frequencies at the price of some added complication in the model. Examples are the naïve Bayes multinomial model (MM), the Dirichlet compound multinomial model (DCM) and the Exponential-Family Approximation of the DCM (EDCM), as well as support vector machines (SVM). Although it is usually claimed that incorporating local word frequency in a document improves text classification performance, we here test whether such claims are true or not. In this paper we show experimentally that simplified forms of the MM, EDCM, and SVM models which ignore the frequency of each word in a document perform about at the same level as MM, DCM, EDCM and SVM models which incorporate local term frequency. We also present a new form of the naïve Bayes multivariate Bernoulli model (MBM) which is able to make use of local term frequency and show again that it offers no significant advantage over the plain MBM. We conclude that word burstiness is so strong that additional occurrences of a word essentially add no useful information to a classifier.

Transaction logs from online search engines are valuable for two reasons: First, they provide insight into human information-seeking behavior. Second, log data can be used to train user models, which can then be applied to improve retrieval systems. This article presents a study of logs from PubMed((R)), the public gateway to the MEDLINE((R)) database of bibliographic records from the medical and biomedical primary literature. Unlike most previous studies on general Web search, our work examines user activities with a highly-specialized search engine. We encode user actions as string sequences and model these sequences using n-gram language models. The models are evaluated in terms of perplexity and in a sequence prediction task. They help us better understand how PubMed users search for information and provide an enabler for improving users' search experience.

It is known that users of internet search engines often enter queries with misspellings in one or more search terms. Several web search engines make suggestions for correcting misspelled words, but the methods used are proprietary and unpublished to our knowledge. Here we describe the methodology we have developed to perform spelling correction for the PubMed search engine. Our approach is based on the noisy channel model for spelling correction and makes use of statistics harvested from user logs to estimate the probabilities of different types of edits that lead to misspellings. The unique problems encountered in correcting search engine queries are discussed and our solutions are outlined.

Topic Detection and Tracking (TDT) is a research initiative that aims at techniques to organize news documents in terms of news events. We propose a method that incorporates simple semantics into TDT by splitting the term space into groups of terms that have the meaning of the same type. Such a group can be associated with an external ontology. This ontology is used to determine the similarity of two terms in the given group. We extract proper names, locations, temporal expressions and normal terms into distinct sub-vectors of the document representation. Measuring the similarity of two documents is conducted by comparing a pair of their corresponding sub-vectors at a time. We use a simple perceptron to optimize the relative emphasis of each semantic class in the tracking and detection decisions. The results suggest that the spatial and the temporal similarity measures need to be improved. Especially the vagueness of spatial and temporal terms needs to be addressed.

This paper investigates the effectiveness of using MeSH(®) in PubMed through its automatic query expansion process: Automatic Term Mapping (ATM). We run Boolean searches based on a collection of 64 topics and about 160,000 MEDLINE(®) citations used in the 2006 and 2007 TREC Genomics Tracks. For each topic, we first automatically construct a query by selecting keywords from the question. Next, each query is expanded by ATM, which assigns different search tags to terms in the query. Three search tags: [MeSH Terms], [Text Words], and [All Fields] are chosen to be studied after expansion because they all make use of the MeSH field of indexed MEDLINE citations. Furthermore, we characterize the two different mechanisms by which the MeSH field is used. Retrieval results using MeSH after expansion are compared to those solely based on the words in MEDLINE title and abstracts. The aggregate retrieval performance is assessed using both F-measure and mean rank precision. Experimental results suggest that query expansion using MeSH in PubMed can generally improve retrieval performance, but the improvement may not affect end PubMed users in realistic situations.

Concept based video retrieval often relies on imperfect and uncertain concept detectors. We propose a general ranking framework to define effective and robust ranking functions, through explicitly addressing detector uncertainty. It can cope with multiple concept-based representations per video segment and it allows the re-use of effective text retrieval functions which are defined on similar representations. The final ranking status value is a weighted combination of two components: the expected score of the possible scores, which represents the risk-neutral choice, and the scores' standard deviation, which represents the risk or opportunity that the score for the actual representation is higher. The framework consistently improves the search performance in the shot retrieval task and the segment retrieval task over several baselines in five TRECVid collections and two collections which use simulated detectors of varying performance.

Different term weighting techniques such as [Formula: see text] or have been used intensely for manifold text-based information retrieval tasks. Their use for modeling term profiles for named entities and subsequent calculation of similarities between these named entities have been studied to a much smaller extent. The recent trend of microblogging made available massive amounts of information about almost every topic around the world. Therefore, microblogs represent a valuable source for text-based named entity modeling. In this paper, we present a systematic and comprehensive evaluation of different , , , , and for the task of inferring similarities between named entities, based on data extracted from . We analyze several thousand combinations of choices for the above mentioned dimensions, which influence the similarity calculation process, and we investigate in which way they impact the quality of the similarity estimates. Evaluation is performed using three real-world data sets: two collections of microblogs related to music artists and one related to movies. For the music collections, we present results of using as benchmark genre information from allmusic.com. For the movie collection, we present results of using as benchmark categories from IMDb. We show that microblogs can indeed be exploited to model named entity similarity with remarkable accuracy, provided the correct settings for the analyzed aspects are used. We further compare the results to those obtained when using Web pages as data source.

Most of the fastest-growing string collections today are repetitive, that is, most of the constituent documents are similar to many others. As these collections keep growing, a key approach to handling them is to exploit their repetitiveness, which can reduce their space usage by orders of magnitude. We study the problem of indexing repetitive string collections in order to perform efficient document retrieval operations on them. Document retrieval problems are routinely solved by search engines on large natural language collections, but the techniques are less developed on generic string collections. The case of repetitive string collections is even less understood, and there are very few existing solutions. We develop two novel ideas, and , that yield highly compressed indexes solving the problem of document listing (find all the documents where a string appears), top- document retrieval (find the documents where a string appears most often), and document counting (count the number of documents where a string appears). We also show that a classical data structure supporting the latter query becomes highly compressible on repetitive data. Finally, we show how the tools we developed can be combined to solve ranked conjunctive and disjunctive multi-term queries under the simple [Formula: see text] model of relevance. We thoroughly evaluate the resulting techniques in various real-life repetitiveness scenarios, and recommend the best choices for each case.

Although youth are increasingly going online to fulfill their needs for information, many youth struggle with information and digital literacy skills, such as the abilities to conduct a search and assess the credibility of online information. Ideally, these skills encompass an accurate and comprehensive understanding of the ways in which a system, such as a Web search engine, functions. In order to investigate youths' conceptions of the Google search engine, a drawing activity was conducted with 26 after-school program participants to elicit their mental models of Google. The findings revealed that many participants personified Google and emphasized anthropomorphic elements, computing equipment, and/or connections (such as cables, satellites and antennas) in their drawings. Far fewer participants focused their drawings on the actual Google interface or on computer code. Overall, their drawings suggest a limited understanding of Google and the ways in which it actually works. However, an understanding of youths' conceptions of Google can enable educators to better tailor their digital literacy instruction efforts and can inform search engine developers and search engine interface designers in making the inner workings of the engine more transparent and their output more trustworthy to young users. With a better understanding of how Google works, young users will be better able to construct effective queries, assess search results, and ultimately find relevant and trustworthy information that will be of use to them.

Recommender systems have dramatically changed the way we consume content. Internet applications rely on these systems to help users navigate among the ever-increasing number of choices available. However, most current systems ignore the fact that user preferences can change according to context, resulting in recommendations that do not fit user interests. This research addresses these issues by proposing the architecture, which uses local learning techniques to embed contextual awareness into collaborative filtering models. The proposed architecture is demonstrated on two large-scale case studies involving over 130 million and over 7 million unique samples, respectively. Results show that contextual models trained with a small fraction of the data provided similar accuracy to collaborative filtering models trained with the complete dataset. Moreover, the impact of taking into account context in real-world datasets has been demonstrated by higher accuracy of context-based models in comparison to random selection models.

Every information retrieval (IR) model embeds in its scoring function a form of term frequency (TF) quantification. The contribution of the term frequency is determined by the properties of the function of the chosen TF quantification, and by its TF normalization. The first defines how independent the occurrences of multiple terms are, while the second acts on mitigating the a priori probability of having a high term frequency in a document (estimation usually based on the document length). New test collections, coming from different domains (e.g. medical, legal), give evidence that not only document length, but in addition, verboseness of documents should be explicitly considered. Therefore we propose and investigate a systematic combination of document verboseness and length. To theoretically justify the combination, we show the duality between document verboseness and length. In addition, we investigate the duality between verboseness and other components of IR models. We test these new TF normalizations on four suitable test collections. We do this on a well defined spectrum of TF quantifications. Finally, based on the theoretical and experimental observations, we show how the two components of this new normalization, document verboseness and length, interact with each other. Our experiments demonstrate that the new models never underperform existing models, while sometimes introducing statistically significantly better results, at no additional computational cost.

The humanities, like many other areas of society, are currently undergoing major changes in the wake of digital transformation. However, in order to make collection of digitised material in this area easily accessible, we often still lack adequate search functionality. For instance, digital archives for textiles offer keyword search, which is fairly well understood, and arrange their content following a certain taxonomy, but search functionality at the level of thread structure is still missing. To facilitate the clustering and search, we introduce an approach for recognising similar weaving patterns based on their structures for textile archives. We first represent textile structures using hypergraphs and extract multisets of -neighbourhoods describing weaving patterns from these graphs. Then, the resulting multisets are clustered using various distance measures and various clustering algorithms (K-Means for simplicity and hierarchical agglomerative algorithms for precision). We evaluate the different variants of our approach experimentally, showing that this can be implemented efficiently (meaning it has linear complexity), and demonstrate its quality to query and cluster datasets containing large textile samples. As, to the best of our knowledge, this is the first practical approach for explicitly modelling complex and irregular weaving patterns usable for retrieval, we aim at establishing a solid baseline.

Pretrained multilingual text encoders based on neural , such as multilingual BERT (mBERT) and XLM, have recently become a default paradigm for cross-lingual transfer of natural language processing models, rendering cross-lingual word embedding spaces (CLWEs) effectively obsolete. In this work we present a systematic empirical study focused on the suitability of the state-of-the-art multilingual encoders for cross-lingual document and sentence retrieval tasks across a number of diverse language pairs. We first treat these models as multilingual text encoders and benchmark their performance in unsupervised ad-hoc sentence- and document-level CLIR. In contrast to supervised language understanding, our results indicate that for unsupervised document-level CLIR-a setup with no relevance judgments for IR-specific fine-tuning-pretrained multilingual encoders on average fail to significantly outperform earlier models based on CLWEs. For sentence-level retrieval, we do obtain state-of-the-art performance: the peak scores, however, are met by multilingual encoders that have been further specialized, in a supervised fashion, for sentence understanding tasks, rather than using their vanilla 'off-the-shelf' variants. Following these results, we introduce localized relevance matching for document-level CLIR, where we independently score a query against document sections. In the second part, we evaluate multilingual encoders fine-tuned in a supervised fashion (i.e., we ) on English relevance data in a series of zero-shot language and domain transfer CLIR experiments. Our results show that, despite the supervision, and due to the domain and language shift, supervised re-ranking rarely improves the performance of multilingual transformers as unsupervised base rankers. Finally, only with in-domain contrastive fine-tuning (i.e., same domain, only language transfer), we manage to improve the ranking quality. We uncover substantial empirical differences between cross-lingual retrieval results and results of (zero-shot) cross-lingual transfer for monolingual retrieval in target languages, which point to "monolingual overfitting" of retrieval models trained on monolingual (English) data, even if they are based on multilingual transformers.