This study delves into the mechanisms that spark user curiosity driving active engagement within public Telegram groups. By analyzing approximately 6 million messages from 29,196 users across 409 groups, we identify and quantify the key factors that stimulate users to actively participate (i.e., send messages) in group discussions. These factors include social influence, novelty, complexity, uncertainty, and conflict, all measured through metrics derived from message sequences and user participation over time. After clustering the messages, we apply explainability techniques to assign meaningful labels to the clusters. This approach uncovers macro categories representing distinct curiosity stimulation profiles, each characterized by a unique combination of various stimuli. Social influence from peers and influencers drives engagement for some users, while for others, rare media types or a diverse range of senders and media sparks curiosity. Analyzing patterns, we found that user curiosity stimuli are mostly stable, but, as the time between the initial message increases, curiosity occasionally shifts. A graph-based analysis of influence networks reveals that users motivated by direct social influence tend to occupy more peripheral positions, while those who are not stimulated by any specific factors are often more central, potentially acting as initiators and conversation catalysts. These findings contribute to understanding information dissemination and spread processes on social media networks, potentially contributing to more effective communication strategies.

We tackle Extreme Multi-Label Text Classification (XMTC), which involves assigning relevant labels to texts from a huge label space. Attempting to optimize the underexplored tail-head trade-off, we address the XMTC task through its core challenges of volume, skewness, and quality by proposing xCoRetriev, a novel two-stage retrieving and fusing ranking pipeline. Our pipeline addresses the volume challenge by dynamically slicing the large label space; it also tackles the skewness challenge by favoring the tail labels while fusing sparse and dense retrievers. Finally, xCoRetriev faces the quality challenge by enhancing the label space with Retrieval-Augmented Generated (RAG)-labels. Our experiments with four XMTC benchmarks with hundreds of thousands of text documents and labels against six state-of-the-art XMTC baselines demonstrate xCoRetriev's strengths in terms of: (i)~scalability for large label spaces, being among the most efficient methods at training and prediction; (ii)~effectiveness in the face of high skewness, with gains of up to 48% in propensity-scored metrics against the best state-of-the-art baselines; and (iii)~capability of handling very noisy datasets by exploiting RAG-labels.

Fine-tuning transformer-based deep-learning models are currently at the forefront of natural language processing (NLP) and information retrieval (IR) tasks. However, fine-tuning these transformers for specific tasks, especially when dealing with ever-expanding volumes of data, constant retraining requirements, and budget constraints, can be computationally and financially costly, requiring substantial energy consumption and contributing to carbon dioxide emissions. This article focuses on advancing the state-of-the-art (SOTA) on instance selection (IS)—a range of document filtering techniques designed to select the most representative documents for the sake of training. The objective is to either maintain or enhance classification effectiveness while reducing the overall training (fine-tuning) total processing time. In our prior research, we introduced the E2SC framework, a redundancy-oriented IS method focused on transformers and large datasets—currently the state-of-the-art in IS. Nonetheless, important research questions remained unanswered in our previous work, mostly due to E2SC’s sole emphasis on redundancy. In this article, we take our research a step further by proposing biO-IS— an extended bi - o bjective i nstance s election solution, a novel IS framework aimed at simultaneously removing redundant and noisy instances from the training. biO-IS estimates redundancy based on scalable, fast, and calibrated weak classifiers and captures noise with the support of a new entropy-based step. We also propose a novel iterative process to estimate near-optimum reduction rates for both steps. Our extended solution is able to reduce the training sets by 41% on average (up to 60%) while maintaining the effectiveness in all tested datasets, with speedup gains of 1.67 on average (up to 2.46x). No other baseline, not even our previous SOTA solution, was capable of achieving results with this level of quality, considering the tradeoff among training reduction, effectiveness, and speedup. To ensure reproducibility, our documentation, code, and datasets can be accessed on GitHub— https://github.com/waashk/bio-is .

Transformer-based deep learning is currently the state-of-the-art in many NLP and IR tasks. However, fine-tuning such Transformers for specific tasks, especially in scenarios of ever-expanding volumes of data with constant re-training requirements and budget constraints, is costly (computationally and financially) and energy-consuming. In this paper, we focus on Instance Selection (IS) - a set of methods focused on selecting the most representative documents for training, aimed at maintaining (or improving) classification effectiveness while reducing total time for training (or fine-tuning). We propose E2SC-IS -- Effective, Efficient, and Scalable Confidence-Based IS -- a two-step framework with a particular focus on Transformers and large datasets. E2SC-IS estimates the probability of each instance being removed from the training set based on scalable, fast, and calibrated weak classifiers. E2SC-IS also exploits iterative heuristics to estimate a near-optimal reduction rate. Our solution can reduce the training sets by 29% on average while maintaining the effectiveness in all datasets, with speedup gains up to 70%, scaling for very large datasets (something that the baselines cannot do).

Record Deduplication (RD) aims to identify instances that represent the same real-world entity in data repositories. In the government environment, the RD process facilitates the identification of irregularities and reduces the consumption of computing resources in data integration tasks. In this context, we propose a scalable, effective and efficient platform, called DedupeGov, for integrating large data repositories (i.e., with large volumes of data, in the order of millions of records) to unify duplicate entities from multiple and different sources. Our experimental results indicate a 21.8% of reduction in the number of records of the original repository with 99% of precision and 95% of recall when identifying duplicate records. In addition, our platform was capable of building more complete records, eliminating at least 32% of records with null attributes. Furthermore, our solution is very efficient and scalable for large volumes of data, deduplicating a repository of almost 400 million records in around one hour, besides being easy to generalize to different types of entity.

Learning to Rank (L2R) is the core task of many Information Retrieval systems. Recently, a great effort has been put on exploring Deep Neural Networks (DNNs) for L2R, with significant results. However, risk-sensitiveness, an important and recent advance in the L2R arena, that reduces variability and increases trust, has not been incorporated into Deep Neural L2R yet. Risk-sensitive measures are important to assess the risk of an IR system to perform worse than a set of baseline IR systems for several queries. However, the risk-sensitive measures described in the literature have a non-smooth behavior, making them difficult, if not impossible, to be optimized by DNNs. In this work we solve this difficult problem by proposing a family of new loss functions -- \riskloss\ -- that support a smooth risk-sensitive optimization. \riskloss\ introduces two important contributions: (i) the substitution of the traditional NDCG or MAP metrics in risk-sensitive measures with smooth loss functions that evaluate the correlation between the predicted and the true relevance order of documents for a given query and (ii) the use of distinct versions of the same DNN architecture as baselines by means of a multi-dropout technique during the smooth risk-sensitive optimization, avoiding the inconvenience of assessing multiple IR systems as part of DNN training. We empirically demonstrate significant achievements of the proposed \riskloss\ functions when used with recent DNN methods in the context of well-known web-search datasets such as WEB10K, YAHOO, and MQ2007. Our solutions reach improvements of 8% in effectiveness (NDCG) while improving in around 5% the risk-sensitiveness (\grisk\ measure) when applied together with a state-of-the-art Self-Attention DNN-L2R architecture. Furthermore, \riskloss\ is capable of reducing by 28% the losses over the best evaluated baselines and significantly improving over the risk-sensitive state-of-the-art non-DNN method (by up to 13.3%) while keeping (or even increasing) overall effectiveness. All these results ultimately establish a new level for the state-of-the-art on risk-sensitiveness and DNN-L2R research.

Attention is a scarce resource disputed by algorithms and people on the Internet. This competition for attention is part of online spaces especially online small groups where there is a limited number of individuals interacting with each other using text and media content that is not controlled by algorithms or human curators. In these groups, as certain participants and piece of content can catch the collective attention, a question that naturally arises is: how to analyze topic attention in online small groups? In this paper, we propose a methodology aimed at answering this question. Our proposal consists of sets of analyses over topical (obtained from topic analysis) transition graphs for characterizing attention allocation, permanence and shifting as well as participant role characterization during discussions in online small groups. We experimented with our methodology using WhatsApp groups as a case study. Among other results, we identified and characterized abrupt and smooth topic transitions as well as patterns of participant activity related to certain topics.

Recent advances in text-related tasks on the Web, such as text (topic) classification and sentiment analysis, have been made possible by exploiting mostly the "rule of more": more data (massive amounts) more computing power, more complex solutions. We propose a shift in the paradigm to do "more with less" by focusing, at maximum extent, just on the task at hand (e.g., classify a single test instance). Accordingly, we propose MetaLazy, a new supervised lazy text classification meta-strategy that greatly extends the scope of lazy solutions. Lazy classifiers postpone the creation of a classification model until a given test instance for decision making is given. MetaLazy exploits new ideas and solutions, which have in common their lazy nature, producing altogether a solution for text classification, which is simpler, more efficient, and less data demanding than new alternatives. It extends and evolves the lazy creation of the model for the test instance by allowing: (i) to dynamically choose the best classifier for the task; (ii) the exploration of distances in the neighborhood of the test document when learning a classification model, thus diminishing the importance of irrelevant training instances; and (iii) a better representational space for training and test documents by augmenting them, in a lazy fashion, with new co-occurrence based features considering just those observed in the specific test instance. In a sizeable experimental evaluation, considering topics and sentiment analysis datasets and nine baselines, we show that our MetaLazy instantiations are among the top performers in most situations, even when compared to state-of-the-art deep learning classifiers such as Deep Network Transformer Architectures.

We propose new solutions that enhance and extend the already very successful application of meta-features to text classification. Our newly proposed meta-features are capable of: (1) improving the correlation of small pieces of evidence shared by neighbors with labeled categories by means of synthetic document representations and (local and global) hyperplane distances; and (2) estimating the level of error introduced by these newly proposed and the existing meta-features in the literature, specially for hard-to-classify regions of the feature space. Our experiments with large and representative number of datasets show that our new solutions produce the best results in all tested scenarios, achieving gains of up to 12% over the strongest meta-feature proposal of the literature.

In this paper, we advance the state-of-the-art in topic modeling by means of a new document representation based on pre-trained word embeddings for non-probabilistic matrix factorization. Specifically, our strategy, called CluWords, exploits the nearest words of a given pre-trained word embedding to generate meta-words capable of enhancing the document representation, in terms of both, syntactic and semantic information. The novel contributions of our solution include: (i)the introduction of a novel data representation for topic modeling based on syntactic and semantic relationships derived from distances calculated within a pre-trained word embedding space and (ii)the proposal of a new TF-IDF-based strategy, particularly developed to weight the CluWords. In our extensive experimentation evaluation, covering 12 datasets and 8 state-of-the-art baselines, we exceed (with a few ties) in almost cases, with gains of more than 50% against the best baselines (achieving up to 80% against some runner-ups). Finally, we show that our method is able to improve document representation for the task of automatic text classification.

Text representation models are the fundamental basis for information retrieval and text mining tasks. Although different text models have been proposed, they typically target specific task aspects in isolation, such as time efficiency, accuracy, or applicability for different scenarios. Here we present Bag of Textual Graphs (BoTG), a general text representation model that addresses these three requirements at the same time. The proposed textual representation is based on a graph‐based scheme that encodes term proximity and term ordering, and represents text documents into an efficient vector space that addresses all these aspects as well as provides discriminative textual patterns. Extensive experiments are conducted in two experimental scenarios—classification and retrieval—considering multiple well‐known text collections. We also compare our model against several methods from the literature. Experimental results demonstrate that our model is generic enough to handle different tasks and collections. It is also more efficient than the widely used state‐of‐the‐art methods in textual classification and retrieval tasks, with a competitive effectiveness, sometimes with gains by large margins.

Sentiment analysis has become a very important tool for analysis of social media data. There are several methods developed, covering distinct aspects of the problem and disparate strategies. However, no single technique fits well in all cases or for all data sources. Supervised approaches may be able to adapt to specific situations, but require manually labeled training, which is very cumbersome and expensive to acquire, mainly for a new application. In this context, we propose to combine several popular and effective state‐of‐the‐practice sentiment analysis methods by means of an unsupervised bootstrapped strategy. One of our main goals is to reduce the large variability (low stability) of the unsupervised methods across different domains. The experimental results demonstrate that our combined method (aka, 10SENT) improves the effectiveness of the classification task, considering thirteen different data sets. Also, it tackles the key problem of cross‐domain low stability and produces the best (or close to best) results in almost all considered contexts, without any additional costs (e.g., manual labeling). Finally, we also investigate a transfer learning approach for sentiment analysis to gather additional (unsupervised) information for the proposed approach, and we show the potential of this technique to improve our results.

Learning to Rank (L2R) is one of the main research lines in Information Retrieval. Risk-sensitive L2R is a sub-area of L2R that tries to learn models that are good on average while at the same time reducing the risk of performing poorly in a few but important queries (e.g., medical or legal queries). One way of reducing risk in learned models is by selecting and removing noisy, redundant features, or features that promote some queries to the detriment of others. This is exacerbated by learning methods that usually maximize an average metric (e.g., mean average precision (MAP) or Normalized Discounted Cumulative Gain (NDCG)). However, historically, feature selection (FS) methods have focused only on effectiveness and feature reduction as the main objectives. Accordingly, in this work, we propose to evaluate FS for L2R with an additional objective in mind, namely risk-sensitiveness . We present novel single and multi-objective criteria to optimize feature reduction, effectiveness, and risk-sensitiveness, all at the same time. We also introduce a new methodology to explore the search space, suggesting effective and efficient extensions of a well-known Evolutionary Algorithm (SPEA2) for FS applied to L2R. Our experiments show that explicitly including risk as an objective criterion is crucial to achieving a more effective and risk-sensitive performance. We also provide a thorough analysis of our methodology and experimental results.

In this paper, we advance the state-of-the-art in topic modeling by means of the design and development of a novel (semi-formal) general topic modeling framework. The novel contributions of our solution include: (i) the introduction of new semantically-enhanced data representations for topic modeling based on pooling, and (ii) the proposal of a novel topic extraction strategy - ASToC - that solves the difficulty in representing topics in our semantically-enhanced information space. In our extensive experimentation evaluation, covering 12 datasets and 12 state-of-the-art baselines, totalizing 108 tests, we exceed (with a few ties) in almost 100 cases, with gains of more than 50% against the best baselines (achieving up to 80% against some runner-ups). We provide qualitative and quantitative statistical analyses of why our solutions work so well. Finally, we show that our method is able to improve document representation in automatic text classification.

In this paper, we propose a general framework that allows developers to filter, summarize and analyze user reviews written about applications on App Stores. Our framework extracts automatically relevant features from reviews of apps (e.g., information about functionalities, bugs, requirements, etc) and analyzes the sentiment associated with each of them. Our framework has three main building blocks, namely, (i) topic modeling, (ii) sentiment analysis and (iii) summarization interface. The topic modeling block aims at finding semantic topics from textual comments, extracting the target features based on the most relevant words of each discovered topic. The sentiment analysis block detects the sentiment associated with each discovered feature. The summarization interface provides to developers an intuitive visualization of the features (i.e., topics) and their associated sentiment, providing richer information than a 'star rating' strategy. Our evaluation shows that the topic modeling block is able to organize information provided by users in subcategories that facilitate the understanding of which features more positively/negatively impact the overall evaluation of the application. Regarding user satisfaction, we can observe that, in spite of the star rating being a good measure of evaluation, the Sentiment Analysis technique is more accurate in capturing the sentiment transmitted by the user by means of a comment.

We address the problem of automatically learning to classify texts by exploiting information derived from meta-features, i.e., features derived from the original bag-of-words representation. Specifically, we provide an in-depth analysis on the recently proposed distance-based meta-features, a data engineering technique that relies on the distance between documents to transform the original feature space into a new one, potentially smaller and more informed. Despite its potential, the meta-feature space may be unnecessarily complex and highly dimensional, which increases the tendency of overfitting, limits the application of meta-features in different contexts, and increases computational costs. In this work, we propose the use of multi-objective strategies to reduce the number of meta-features while maximizing the classification effectiveness, when considering the adequacy of the selected meta-features to a particular dataset or classification method. We present effective and efficient proposals for meta-feature selection that can substantially reduce the number of meta-features by up to 89 percent while keeping or improving the classification effectiveness, something not possible with any of the evaluated baselines. We also use our selection strategies as evaluation tools to analyze different combinations of meta-features. We found very compact combinations of meta-features that can achieve high classification effectiveness in most datasets, despite their peculiarities.

Random Forest (RF) is one of the most successful strategies for automated classification tasks. Motivated by the RF success, recently proposed RF-based classification approaches leverage the central RF idea of aggregating a large number of low-correlated trees, which are inherently parallelizable and provide exceptional generalization capabilities. In this context, this work brings several new contributions to this line of research. First, we propose a new RF-based strategy (BERT) that applies the boosting technique in bags of extremely randomized trees. Second, we empirically demonstrate that this new strategy, as well as the recently proposed BROOF and LazyNN_RF classifiers do complement each other, motivating us to stack them to produce an even more effective classifier. Up to our knowledge, this is the first strategy to effectively combine the three main ensemble strategies: stacking, bagging (the cornerstone of RFs) and boosting. Finally, we exploit the efficient and unbiased stacking strategy based on out-of-bag (OOB) samples to considerably speedup the very costly training process of the stacking procedure. Our experiments in several datasets covering two high-dimensional and noisy domains of topic and sentiment classification provide strong evidence in favor of the benefits of our RF-based solutions. We show that BERT is among the top performers in the vast majority of analyzed cases, while retaining the unique benefits of RF classifiers (explainability, parallelization, easiness of parameterization). We also show that stacking only the recently proposed RF-based classifiers and BERT using our OOB-based strategy is not only significantly faster than recently proposed stacking strategies (up to six times) but also much more effective, with gains up to 21% and 17% on MacroF1 and MicroF1, respectively, over the best base method, and of 5% and 6% over a stacking of traditional methods, performing no worse than a complete stacking of methods at a much lower computational effort.

Tags (keywords freely assigned by users to describe web content) have become highly popular on Web 2.0 applications, because of the strong stimuli and easiness for users to create and describe their own content. This increase in tag popularity has led to a vast literature on tag recommendation methods. These methods aim at assisting users in the tagging process, possibly increasing the quality of the generated tags and, consequently, improving the quality of the information retrieval (IR) services that rely on tags as data sources. Regardless of the numerous and diversified previous studies on tag recommendation, to our knowledge, no previous work has summarized and organized them into a single survey article. In this article, we propose a taxonomy for tag recommendation methods, classifying them according to the target of the recommendations, their objectives, exploited data sources, and underlying techniques. Moreover, we provide a critical overview of these methods, pointing out their advantages and disadvantages. Finally, we describe the main open challenges related to the field, such as tag ambiguity, cold start, and evaluation issues.

User‐generated content is one of the most interesting phenomena of current published media, as users are now able not only to consume, but also to produce content in a much faster and easier manner. However, such freedom also carries concerns about content quality. In this work, we propose an automatic framework to assess the quality of collaboratively generated content. Quality is addressed as a multidimensional concept, modeled as a combination of independent assessments, each regarding different quality dimensions. Accordingly, we adopt a machine‐learning (ML)‐based multiview approach to assess content quality. We perform a thorough analysis of our framework on two different domains: Questions and Answer Forums and Collaborative Encyclopedias. This allowed us to better understand when and how the proposed multiview approach is able to provide accurate quality assessments. Our main contributions are: (a) a general ML multiview framework that takes advantage of different views of quality indicators; (b) the improvement (up to 30%) in quality assessment over the best state‐of‐the‐art baseline methods; (c) a thorough feature and view analysis regarding impact, informativeness, and correlation, based on two distinct domains.

The vast majority of the current author name disambiguation solutions are designed to disambiguate a whole digital library (DL) at once considering the entire repository. However, these solutions besides being very expensive and having scalability problems, also may not benefit from eventual manual corrections, as they may be lost whenever the process of disambiguating the entire repository is required. In the real world, in which repositories are updated on a daily basis, incremental solutions that disambiguate only the newly introduced citation records, are likely to produce improved results in the long run. However, the problem of incremental author name disambiguation has been largely neglected in the literature. In this article we present a new author name disambiguation method, specially designed for the incremental scenario. In our experiments, our new method largely outperforms recent incremental proposals reported in the literature as well as the current state‐of‐the‐art non‐incremental method.

Learning to rank (L2R) algorithms use a labeled training set to generate a ranking model that can be later used to rank new query results. These training sets are very costly and laborious to produce, requiring human annotators to assess the relevance or order of the documents in relation to a query. Active learning (AL) algorithms are able to reduce the labeling effort by actively sampling an unlabeled set and choosing data instances that maximize the effectiveness of a learning function. But AL methods require constant supervision, as documents have to be labeled at each round of the process. In this paper, we propose that certain characteristics of unlabeled L2R datasets allow for an unsupervised, compression-based selection process to be used to create small and yet highly informative and effective initial sets that can later be labeled and used to bootstrap a L2R system. We implement our ideas through a novel unsupervised selective sampling method, which we call Cover, that has several advantages over AL methods tailored to L2R. First, it does not need an initial labeled seed set and can select documents from scratch. Second, selected documents do not need to be labeled as the iterations of the method progress since it is unsupervised (i.e., no learning model needs to be updated). Thus, an arbitrarily sized training set can be selected without human intervention depending on the available budget. Third, the method is efficient and can be run on unlabeled collections containing millions of query-document instances. We run various experiments with two important L2R benchmarking collections to show that the proposed method allows for the creation of small, yet very effective training sets. It achieves full training-like performance with less than 10% of the original sets selected, outperforming the baselines in both effectiveness and scalability.

Learning to Rank (L2R) is currently an essential task in basically all types of information systems given the huge and ever increasing amount of data made available. While many solutions have been proposed to improve L2R functions, relatively little attention has been paid to the task of improving the quality of the feature space. L2R strategies usually rely on dense feature representations, which contain noisy or redundant features, increasing the cost of the learning process, without any benefits. Although feature selection (FS) strategies can be applied to reduce dimensionality and noise, side effects of such procedures have been neglected, such as the risk of getting very poor predictions in a few (but important) queries. In this paper we propose multi-objective FS strategies that optimize both aspects at the same time: ranking performance and risk-sensitive evaluation. For this, we approximate the Pareto-optimal set for multi-objective optimization in a new and original application to L2R. Our contributions include novel FS methods for L2R which optimize multiple, potentially conflicting, criteria. In particular, one of the objectives (risk-sensitive evaluation) has never been optimized in the context of FS for L2R before. Our experimental evaluation shows that our proposed methods select features that are more effective (ranking performance) and low-risk than those selected by other state-of-the-art FS methods.

Record deduplication aims at identifying entities that are potentially the same in a data repository. A set of pairs that is manually labeled is generally used to tune the deduplication process, as each dataset has a particular dirtiness pattern. However, producing an informative set of pairs is a very costly task, especially in very large datasets (even for expert users). We propose a new sampling strategy that is able to select a very small and informative set of pairs from large datasets. Our results show that our approach reduces user effort substantially while achieving a competitive or superior matching quality.

Expertise retrieval has been the subject of intense research over the past decade, particularly with the public availability of benchmark test collections for expertise retrieval in enterprises. Another domain which has seen comparatively less research on expertise retrieval is academic search. In this paper, we describe the Lattes Expertise Retrieval (LExR) test collection for research on academic expertise retrieval. LExR has been designed to provide a large-scale benchmark for two complementary expertise retrieval tasks, namely, expert profiling and expert finding. Unlike currently available test collections, which fully support only one of these tasks, LExR provides graded relevance judgments performed by expert judges separately for each task. In addition, LExR is both cross-organization and cross-area, encompassing candidate experts from all areas of knowledge working in research institutions all over Brazil. As a result, it constitutes a valuable resource for fostering new research directions on expertise retrieval in an academic setting.

The task of retrieving information that really matters to the users is considered hard when taking into consideration the current and increasingly amount of available information. To improve the effectiveness of this information seeking task, systems have relied on the combination of many predictors by means of machine learning methods, a task also known as learning to rank (L2R). The most effective learning methods for this task are based on ensembles of tress (e.g., Random Forests) and/or boosting techniques (e.g., RankBoost, MART, LambdaMART). In this paper, we propose a general framework that smoothly combines ensembles of additive trees, specifically Random Forests, with Boosting in a original way for the task of L2R. In particular, we exploit out-of-bag samples as well as a selective weight updating strategy (according to the out-of-bag samples) to effectively enhance the ranking performance. We instantiate such a general framework by considering different loss functions, different ways of weighting the weak learners as well as different types of weak learners. In our experiments our rankers were able to outperform all state-of-the-art baselines in all considered datasets, using just a small percentage of the original training set and faster convergence rates.

In this paper we address the problem of automatically learning to classify the sentiment of short messages/reviews by exploiting information derived from meta-level features i.e., features derived primarily from the original bag-of-words representation. We propose new meta-level features especially designed for the sentiment analysis of short messages such as: (i) information derived from the sentiment distribution among the k nearest neighbors of a given short test document x, (ii) the distribution of distances of x to their neighbors and (iii) the document polarity of these neighbors given by unsupervised lexical-based methods. Our approach is also capable of exploiting information from the neighborhood of document x regarding (highly noisy) data obtained from 1.6 million Twitter messages with emoticons. The set of proposed features is capable of transforming the original feature space into a new one, potentially smaller and more informed. Experiments performed with a substantial number of datasets (nineteen) demonstrate that the effectiveness of the proposed sentiment-based meta-level features is not only superior to the traditional bag-of-word representation (by up to 16%) but is also superior in most cases to state-of-art meta-level features previously proposed in the literature for text classification tasks that do not take into account some idiosyncrasies of sentiment analysis. Our proposal is also largely superior to the best lexicon-based methods as well as to supervised combinations of them. In fact, the proposed approach is the only one to produce the best results in all tested datasets in all scenarios.

Tag recommendation strategies that exploit term co‐occurrence patterns with tags previously assigned to the target object have consistently produced state‐of‐the‐art results. However, such techniques work only for objects with previously assigned tags. Here we focus on tag recommendation for objects with no tags, a variation of the well‐known \textit{cold start} problem. We start by evaluating state‐of‐the‐art co‐occurrence based methods in cold start. Our results show that the effectiveness of these methods suffers in this situation. Moreover, we show that employing various automatic filtering strategies to generate an initial tag set that enables the use of co‐occurrence patterns produces only marginal improvements. We then propose a new approach that exploits both positive and negative user feedback to iteratively select input tags along with a genetic programming strategy to learn the recommendation function. Our experimental results indicate that extending the methods to include user relevance feedback leads to gains in precision of up to 58% over the best baseline in cold start scenarios and gains of up to 43% over the best baseline in objects that contain some initial tags (i.e., no cold start). We also show that our best relevance‐feedback‐driven strategy performs well even in scenarios that lack user cooperation (i.e., users may refuse to provide feedback) and user reliability (i.e., users may provide the wrong feedback).

Automatic text classification (TC) continues to be a relevant research topic and several TC algorithms have been proposed. However, the majority of TC algorithms assume that the underlying data distribution does not change over time. In this work, we are concerned with the challenges imposed by the temporal dynamics observed in textual data sets. We provide evidence of the existence of temporal effects in three textual data sets, reflected by variations observed over time in the class distribution, in the pairwise class similarities, and in the relationships between terms and classes. We then quantify, using a series of full factorial design experiments, the impact of these effects on four well‐known TC algorithms. We show that these temporal effects affect each analyzed data set differently and that they restrict the performance of each considered TC algorithm to different extents. The reported quantitative analyses, which are the original contributions of this article, provide valuable new insights to better understand the behavior of TC algorithms when faced with nonstatic (temporal) data distributions and highlight important requirements for the proposal of more accurate classification models.

The design and evaluation of tag recommendation methods has historically focused on maximizing the relevance of the suggested tags for a given object, such as a movie or a song. However, relevance by itself may not be enough to guarantee recommendation usefulness. Promoting novelty and diversity in tag recommendation not only increases the chances that the user will select “some” of the recommended tags but also promotes complementary information (i.e., tags), which helps to cover multiple aspects or topics related to the target object. Previous work has addressed the tag recommendation problem by exploiting at most two of the following aspects: (1) relevance, (2) explicit topic diversity, and (3) novelty. In contrast, here we tackle these three aspects conjointly, by introducing two new tag recommendation methods that cover all three aspects of the problem at different levels. Our first method, called Random Forest with topic-related attributes , or RF t , extends a relevance-driven tag recommender based on the Random Forest ( RF ) learning-to-rank method by including new tag attributes to capture the extent to which a candidate tag is related to the topics of the target object. This solution captures topic diversity as well as novelty at the attribute level while aiming at maximizing relevance in its objective function. Our second method, called Explicit Tag Recommendation Diversifier with Novelty Promotion , or xTReND , reranks the recommendations provided by any tag recommender to jointly promote relevance, novelty, and topic diversity. We use RF t as a basic recommender applied before the reranking, thus building a solution that addresses the problem at both attribute and objective levels. Furthermore, to enable the use of our solutions on applications in which category information is unavailable, we investigate the suitability of using latent Dirichlet allocation (LDA) to automatically generate topics for objects. We evaluate all tag recommendation approaches using real data from five popular Web 2.0 applications. Our results show that RF t greatly outperforms the relevance-driven RF baseline in diversity while producing gains in relevance as well. We also find that our new xTReND reranker obtains considerable gains in both novelty and relevance when compared to that same baseline while keeping the same relevance levels. Furthermore, compared to our previous reranker method, xTReD , which does not consider novelty, xTReND is also quite effective, improving the novelty of the recommended tags while keeping similar relevance and diversity levels in most datasets and scenarios. Comparing our two new proposals, we find that xTReND considerably outperforms RF t in terms of novelty and diversity with only small losses (under 4%) in relevance. Overall, considering the trade-off among relevance, novelty, and diversity, our results demonstrate the superiority of xTReND over the baselines and the proposed alternative, RF t . Finally, the use of automatically generated latent topics as an alternative to manually labeled categories also provides significant improvements, which greatly enhances the applicability of our solutions to applications where the latter is not available.

This paper presents an approach to combine rank aggregation techniques using a soft computing technique -- Genetic Programming -- in order to improve the results in Information Retrieval tasks. Previous work shows that by combining rank aggregation techniques in an agglomerative way, it is possible to get better results than with individual methods. However, these works either combine only a small set of lists or are performed in a completely ad-hoc way. Therefore, given a set of ranked lists and a set of rank aggregation techniques, we propose to use a supervised genetic programming approach to search combinations of them that maximize effectiveness in large search spaces. Experimental results conducted using four datasets with different properties show that our proposed approach reaches top performance in most datasets. Moreover, this cross-dataset performance is not matched by any other baseline among the many we experiment with, some being the state-of-the-art in learning-to-rank and in the supervised rank aggregation tasks. We also show that our proposed framework is very efficient, flexible, and scalable.

Automatic Document Classification (ADC) is the basis of many important applications such as spam filtering and content organization. Naive Bayes (NB) approaches are a widely used classification paradigm, due to their simplicity, efficiency, absence of parameters and effectiveness. However, they do not present competitive effectiveness when compared to other modern statistical learning methods, such as SVMs. This is related to some characteristics of real document collections, such as class imbalance, feature sparseness and strong relationships among attributes. In this paper, we investigate whether the relaxation of the NB feature independence assumption (aka, Semi-NB approaches) can improve its effectiveness in large text collections. We propose four new Lazy Semi-NB strategies that exploit different ideas for alleviating the NB independence assumption. By being lazy, our solutions focus only on the most important features to classify a given test document, overcoming some Semi-NB issues when applied to ADC such as bias towards larger classes and overfitting and/or lack of generalization of the models. We demonstrate that our Lazy Semi-NB proposals can produce superior effectiveness when compared to state-of-the-art ADC classifiers such as SVM and KNN. Moreover, to overcome some efficiency issues of combining Semi-NB and lazy strategies, we take advantage of current manycore GPU architectures and present a massively parallelized version of the Semi-NB approaches. Our experimental results show that speedups of up to 63.36 times can be obtained when compared to serial solutions, making our proposals very practical in real-situations.

The unprecedented growth of available data nowadays has stimulated the development of new methods for organizing and extracting useful knowledge from this immense amount of data. Automatic Document Classification (ADC) is one of such methods, that uses machine learning techniques to build models capable of automatically associating documents to well-defined semantic classes. ADC is the basis of many important applications such as language identification, sentiment analysis, recommender systems, spam filtering, among others. Recently, the use of meta-features has been shown to substantially improve the effectiveness of ADC algorithms. In particular, the use of meta-features that make a combined use of local information (through kNN-based features) and global information (through category centroids) has produced promising results. However, the generation of these meta-features is very costly in terms of both, memory consumption and runtime since there is the need to constantly call the kNN algorithm. We take advantage of the current manycore GPU architecture and present a massively parallel version of the kNN algorithm for highly dimensional and sparse datasets (which is the case for ADC). Our experimental results show that we can obtain speedup gains of up to 15x while reducing memory consumption in more than 5000x when compared to a state-of-the-art parallel baseline. This opens up the possibility of applying meta-features based classification in large collections of documents, that would otherwise take too much time or require the use of an expensive computational platform.

Random Forests (RF) and Boosting are two of the most successful supervised learning paradigms for automatic classification. In this work we propose to combine both strategies in order to exploit their strengths while simultaneously solving some of their drawbacks, especially when applied to high-dimensional and noisy classification tasks. More specifically, we propose a boosted version of the RF classifier (BROOF), which fits an additive model composed by several random forests (as weak learners). Differently from traditional boosting methods which exploit the training error estimate, we here use the stronger out-of-bag (OOB) error estimate which is an out-of-the-box estimate naturally produced by the bagging method used in RFs. The influence of each weak learner in the fitted additive model is inversely proportional to their OOB error. Moreover, the probability of selecting an out-of-bag training example is increased if misclassified by the simpler weak learners, in order to enable the boosted model to focus on complex regions of the input space. We also adopt a selective weight updating procedure, whereas only the out-of-bag examples are updated as the boosting iterations go by. This serves the purpose of slowing down the tendency to focus on just a few hard-to-classify examples. By mitigating this undesired bias known to affect boosting algorithms under high dimensional and noisy scenarios - due to both the selective weighting schema and a proper weak-learner effectiveness assessment - we greatly improve classification effectiveness. Our experiments with several datasets in three representative high-dimensional and noisy domains - topic, sentiment and microarray data classification - an up to ten state-of-the-art classifiers (covering almost 500 results), show that BROOF is the only classifier to be among the top performers in all tested datasets from the topic classification domain, and in the vast majority of cases in sentiment and microarray domains, a surprising result given the knowledge that there is no single top-notch classifier for all datasets.

Building expertise profiles is a crucial step towards identifying experts in different knowledge areas. However, summarizing the topics of expertise of a given individual is a challenging task, primarily due to the semi-structured and heterogeneous nature of the documentary evidence available for this task. In this paper, we investigate the suitability of tag recommendation as a mechanism to produce effective expertise profiles. In particular, we perform a large-scale user study with academic experts from different knowledge areas to assess the effectiveness of multiple supervised and unsupervised tag recommendation approaches as well as multiple sources of textual evidence. Our analysis reveals that traditional content-based tag recommenders perform well at identifying expertise-oriented tags, with article keywords being a particularly effective source of evidence across profiles in different knowledge areas and with various levels of sparsity. Moreover, by combining multiple recommenders and sources of evidence as learning signals, we further demonstrate the effectiveness of tag recommendation for expertise profiling.

The data deduplication task has attracted a considerable amount of attention from the research community in order to provide effective and efficient solutions. The information provided by the user to tune the deduplication process is usually represented by a set of manually labeled pairs. In very large datasets, producing this kind of labeled set is a daunting task since it requires an expert to select and label a large number of informative pairs. In this article, we propose a two-stage sampling selection strategy (T3S) that selects a reduced set of pairs to tune the deduplication process in large datasets. T3S selects the most representative pairs by following two stages. In the first stage, we propose a strategy to produce balanced subsets of candidate pairs for labeling. In the second stage, an active selection is incrementally invoked to remove the redundant pairs in the subsets created in the first stage in order to produce an even smaller and more informative training set. This training set is effectively used both to identify where the most ambiguous pairs lie and to configure the classification approaches. Our evaluation shows that T3S is able to reduce the labeling effort substantially while achieving a competitive or superior matching quality when compared with state-of-the-art deduplication methods in large datasets.

This paper addresses the problem of automatically learning to classify texts by exploiting information derived from meta-level features (i.e., features derived from the original bag-of-words representation). We propose new meta-level features derived from the class distribution, the entropy and the within-class cohesion observed in the k nearest neighbors of a given test document x, as well as from the distribution of distances of x to these neighbors. The set of proposed features is capable of transforming the original feature space into a new one, potentially smaller and more informed. Experiments performed with several standard datasets demonstrate that the effectiveness of the proposed meta-level features is not only much superior than the traditional bag-of-word representation but also superior to other state-of-art meta-level features previously proposed in the literature. Moreover, the proposed meta-features can be computed about three times faster than the existing meta-level ones, making our proposal much more scalable. We also demonstrate that the combination of our meta features and the original set of features produce significant improvements when compared to each feature set used in isolation.

We present a novel 3‐step self‐training method for author name disambiguation—SAND (self‐training associative name disambiguator)—which requires no manual labeling, no parameterization (in real‐world scenarios) and is particularly suitable for the common situation in which only the most basic information about a citation record is available (i.e., author names, and work and venue titles). During the first step, real‐world heuristics on coauthors are able to produce highly pure (although fragmented) clusters. The most representative of these clusters are then selected to serve as training data for the third supervised author assignment step. The third step exploits a state‐of‐the‐art transductive disambiguation method capable of detecting unseen authors not included in any training example and incorporating reliable predictions to the training data. Experiments conducted with standard public collections, using the minimum set of attributes present in a citation, demonstrate that our proposed method outperforms all representative unsupervised author grouping disambiguation methods and is very competitive with fully supervised author assignment methods. Thus, different from other bootstrapping methods that explore privileged, hard to obtain information such as self‐citations and personal information, our proposed method produces topnotch performance with no (manual) training data or parameterization and in the presence of scarce information.

Learning to rank (L2R) algorithms use a labeled training set to generate a ranking model that can later be used to rank new query results. These training sets are costly and laborious to produce, requiring human annotators to assess the relevance or order of the documents in relation to a query. Active learning algorithms are able to reduce the labeling effort by selectively sampling an unlabeled set and choosing data instances that maximize a learning function's effectiveness. In this article, we propose a novel two‐stage active learning method for L2R that combines and exploits interesting properties of its constituent parts, thus being effective and practical. In the first stage, an association rule active sampling algorithm is used to select a very small but effective initial training set. In the second stage, a query‐by‐committee strategy trained with the first‐stage set is used to iteratively select more examples until a preset labeling budget is met or a target effectiveness is achieved. We test our method with various LETOR benchmarking data sets and compare it with several baselines to show that it achieves good results using only a small portion of the original training sets.

Tag recommendation approaches have historically focused on maximizing the relevance of the recommended tags for a given object, such as a movie or a song. Nevertheless, different users may be interested in the same object for different reasons---for instance, the Star Wars movies may appeal to both adventure as well as to fantasy movie fans. In this situation, a sensible strategy is to provide a user with diverse recommendations of how to tag the object. In this paper, we address the problem of recommending relevant and diverse tags as a ranking problem. In particular, we propose a novel tag recommendation approach that explicitly takes into account the possible topics (e.g., categories) underlying an object in order to promote tags with high coverage and low redundancy with respect to these topics. We thoroughly evaluate our proposed approach using data collected from two popular Web 2.0 applications, namely, LastFM and MovieLens. Our experimental results attest the effectiveness of our approach at promoting more relevant and diverse tags in contrast to state-of-the-art relevance-based methods as well as a recently proposed method that takes both relevance and diversity into account.

Collaborative web sites, such as collaborative encyclopedias, blogs, and forums, are characterized by a loose edit control, which allows anyone to freely edit their content. As a consequence, the quality of this content raises much concern. To deal with this, many sites adopt manual quality control mechanisms. However, given their size and change rate, manual assessment strategies do not scale and content that is new or unpopular is seldom reviewed. This has a negative impact on the many services provided, such as ranking and recommendation. To tackle with this problem, we propose a learning to rank (L2R) approach for ranking answers in Q&A forums. In particular, we adopt an approach based on Random Forests and represent query and answer pairs using eight different groups of features. Some of these features are used in the Q&A domain for the first time. Our L2R method was trained to learn the answer rating, based on the feedback users give to answers in Q&A forums. Using the proposed method, we were able (i) to outperform a state of the art baseline with gains of up to 21% in NDCG, a metric used to evaluate rankings; we also conducted a comprehensive study of the features, showing that (ii) review and user features are the most important in the Q&A domain although text features are useful for assessing quality of new answers; and (iii) the best set of new features we proposed was able to yield the best quality rankings.

Deduplication is the task of identifying which objects are potentially the same in a data repository. It usually demands user intervention in several steps of the process, mainly to identify some pairs representing matchings and non-matchings. This information is then used to help in identifying other potentially duplicated records. When deduplication is applied to very large datasets, the performance and matching quality depends on expert users to configure the most important steps of the process (e.g., blocking and classification). In this paper, we propose a new framework called FS-Dedup able to help tuning the deduplication process on large datasets with a reduced effort from the user, who is only required to label a small, automatically selected, subset of pairs. FS-Dedup exploits Signature-Based Deduplication (Sig-Dedup) algorithms in its deduplication core. Sig-Dedup is characterized by high efficiency and scalability in large datasets but requires an expert user to tune several parameters. FS-Dedup helps in solving this drawback by providing a framework that does not demand specialized user knowledge about the dataset or thresholds to produce high effectiveness. Our evaluation over large real and synthetic datasets (containing millions of records) shows that FS-Dedup is able to reach or even surpass the maximal matching quality obtained by Sig-Dedup techniques with a reduced manual effort from the user.

Predicting Web content popularity is an important task for supporting the design and evaluation of a wide range of systems, from targeted advertising to effective search and recommendation services. We here present two simple models for predicting the future popularity of Web content based on historical information given by early popularity measures. Our approach is validated on datasets consisting of videos from the widely used YouTube video-sharing portal. Our experimental results show that, compared to a state-of-the-art baseline model, our proposed models lead to significant decreases in relative squared errors, reaching up to 20% reduction on average, and larger reductions (of up to 71%) for videos that experience a high peak in popularity in their early days followed by a sharp decrease in popularity.

We here propose a new method for expanding entity related queries that automatically filters, weights and ranks candidate expasion terms extracted from Wikipedia articles related to the original query. Our method is based on state-of-the-art tag recommendation methods that exploit heuristic metrics to estimate the descriptive capacity of a given term. Originally proposed for the context of tags, we here apply these recommendation methods to weight and rank terms extracted from multiple fields of Wikipedia articles according to their relevance for the article. We evaluate our method comparing it against three state-of-the-art baselines in three collections. Our results indicate that our method outperforms all baselines in all collections, with relative gains in MAP of up to 14% against the best ones.

Several systems that rely on consistent data to offer high-quality services, such as digital libraries and e-commerce brokers, may be affected by the existence of duplicates, quasi replicas, or near-duplicate entries in their repositories. Because of that, there have been significant investments from private and government organizations for developing methods for removing replicas from its data repositories. This is due to the fact that clean and replica-free repositories not only allow the retrieval of higher quality information but also lead to more concise data and to potential savings in computational time and resources to process this data. In this paper, we propose a genetic programming approach to record deduplication that combines several different pieces of evidence extracted from the data content to find a deduplication function that is able to identify whether two entries in a repository are replicas or not. As shown by our experiments, our approach outperforms an existing state-of-the-art method found in the literature. Moreover, the suggested functions are computationally less demanding since they use fewer evidence. In addition, our genetic programming approach is capable of automatically adapting these functions to a given fixed replica identification boundary, freeing the user from the burden of having to choose and tune this parameter.

This work addresses the task of recommending relevant tags to a target object by jointly exploiting three dimensions of the problem: (i) term co-occurrence with tags pre-assigned to the target object, (ii) terms extracted from multiple textual features, and (iii) several metrics of tag relevance. In particular, we propose several new heuristic methods, which extend state-of-the-art strategies by including new metrics that try to capture how accurately a candidate term describes the object's content. We also exploit two learning-to-rank (L2R) techniques, namely RankSVM and Genetic Programming, for the task of generating ranking functions that combine multiple metrics to accurately estimate the relevance of a tag to a given object. We evaluate all proposed methods in various scenarios for three popular Web 2.0 applications, namely, LastFM, YouTube and YahooVideo. We found that our new heuristics greatly outperform the methods on which they are based, producing gains in precision of up to 181%, as well as another state-of-the-art technique, with improvements in precision of up to 40% over the best baseline in any scenario. Further improvements can also be achieved with the new L2R strategies, which have the additional advantage of being quite flexible and extensible to exploit other aspects of the tag recommendation problem.

We present GreenMeter, a tool for assessing the quality and recommending tags for Web 2.0 content. Its goal is to improve tag quality and the effectiveness of various information services (e.g., search, content recommendation) that rely on tags as data sources. We demonstrate an implementation of GreenMeter for the popular Last.fm application.

Web data repositories usually contain references to thousands of real-world entities from multiple sources. It is not uncommon that multiple entities share the same label (polysemes) and that distinct label variations are associated with the same entity (synonyms), which frequently leads to ambiguous interpretations. Further, spelling variants, acronyms, abbreviated forms, and misspellings compound to worsen the problem. Solving this problem requires identifying which labels correspond to the same real-world entity, a process known as entity resolution. One approach to solve the entity resolution problem is to associate an authority identifier and a list of variant forms with each entity—a data structure known as an authority file. In this work, we propose a generic framework for implementing a method for generating authority files. Our method uses information from the Web to improve the quality of the authority file and, because of that, is referred to as WER—Web-based Entity Resolution. Our contribution here is threefold: (a) we discuss how to implement the WER framework, which is flexible and easy to adapt to new domains; (b) we run extended experimentation with our WER framework to show that it outperforms selected baselines; and (c) we compare the results of a specialized solution for author name resolution with those produced by the generic WER framework, and show that the WER results remain competitive.

This work addresses the task of recommending high quality tags by exploiting not only previously assigned tags, but also terms extracted from other textual features (e.g., title and description) associated with the target object.To estimate the quality of a candidate tag recommendation, we use several metrics related to both tag co-occurrence and information quality. We also propose a heuristic function to combine the metrics to produce a final ranking of the recommended tags. We evaluate our heuristic function in various scenarios, for three popular Web 2.0 applications. Our experimental results indicate that our heuristic function significantly outperforms two state-of-the-art tag recommendation algorithms.

Automatic Document Classification (ADC) is still one of the major information retrieval problems. It usually employs a supervised learning strategy, where we first build a classification model using pre-classified documents and then use this model to classify unseen documents. The majority of supervised algorithms consider that all documents provide equally important information. However, in practice, a document may be considered more or less important to build the classification model according to several factors, such as its timeliness, the venue where it was published in, its authors, among others. In this paper, we are particularly concerned with the impact that temporal effects may have on ADC and how to minimize such impact. In order to deal with these effects, we introduce a temporal weighting function (TWF) and propose a methodology to determine it for document collections. We applied the proposed methodology to ACM-DL and Medline and found that the TWF of both follows a lognormal. We then extend three ADC algorithms (namely kNN, Rocchio and Naïve Bayes) to incorporate the TWF. Experiments showed that the temporally-aware classifiers achieved significant gains, outperforming (or at least matching) state-of-the-art algorithms.

Information extraction by text segmentation (IETS) applies to cases in which data values of interest are organized in implicit semi-structured records available in textual sources (e.g. postal addresses, bibliographic information, ads). It is an important practical problem that has been frequently addressed in the recent literature. In this paper we introduce ONDUX (On Demand Unsupervised Information Extraction), a new unsupervised probabilistic approach for IETS. As other unsupervised IETS approaches, ONDUX relies on information available on pre-existing data to associate segments in the input string with attributes of a given domain. Unlike other approaches, we rely on very effective matching strategies instead of explicit learning strategies. The effectiveness of this matching strategy is also exploited to disambiguate the extraction of certain attributes through a reinforcement step that explores sequencing and positioning of attribute values directly learned on-demand from test data, with no previous human-driven training, a feature unique to ONDUX. This assigns to ONDUX a high degree of flexibility and results in superior effectiveness, as demonstrated by the experimental evaluation we report with textual sources from different domains, in which ONDUX is compared with a state-of-art IETS approach.

Abstract Name ambiguity in the context of bibliographic citations is a difficult problem which, despite the many efforts from the research community, still has a lot of room for improvement. In this article, we present a heuristic‐based hierarchical clustering method to deal with this problem. The method successively fuses clusters of citations of similar author names based on several heuristics and similarity measures on the components of the citations (e.g., coauthor names, work title, and publication venue title). During the disambiguation task, the information about fused clusters is aggregated providing more information for the next round of fusion. In order to demonstrate the effectiveness of our method, we ran a series of experiments in two different collections extracted from real‐world digital libraries and compared it, under two metrics, with four representative methods described in the literature. We present comparisons of results using each considered attribute separately (i.e., coauthor names, work title, and publication venue title) with the author name attribute and using all attributes together. These results show that our unsupervised method, when using all attributes, performs competitively against all other methods, under both metrics, loosing only in one case against a supervised method, whose result was very close to ours. Moreover, such results are achieved without the burden of any training and without using any privileged information such as knowing a priori the correct number of clusters.

Abstract In this work, we investigate the problem of using the block structure of Web pages to improve ranking results. Starting with basic intuitions provided by the concepts of term frequency (TF) and inverse document frequency (IDF), we propose nine block‐weight functions to distinguish the impact of term occurrences inside page blocks, instead of inside whole pages. These are then used to compute a modified BM25 ranking function. Using four distinct Web collections, we ran extensive experiments to compare our block‐weight ranking formulas with two other baselines: (a) a BM25 ranking applied to full pages, and (b) a BM25 ranking that takes into account best blocks. Our methods suggest that our block‐weighting ranking method is superior to all baselines across all collections we used and that average gain in precision figures from 5 to 20% are generated.

The growth of popularity of Web 2.0 applications greatly increased the amount of social media content available on the Internet. However, the unsupervised, user-oriented nature of this source of information, and thus, its potential lack of quality, have posed a challenge to information retrieval (IR) services. Previous work focuses mostly only on tags, although a consensus about its effectiveness as supporting information for IR services has not yet been reached. Moreover, other textual features of the Web 2.0 are generally overseen by previous research.

The classification performance of an associative classifier is strongly dependent on the statistic measure or metric that is used to quantify the strength of the association between features and classes (i.e., confidence, correlation etc.). Previous studies have shown that classifiers produced by different metrics may provide conflicting predictions, and that the best metric to use is data-dependent and rarely known while designing the classifier. This uncertainty concerning the optimal match between metrics and problems is a dilemma, and prevents associative classifiers to achieve their maximal performance. This dilemma is the focus of this paper. A possible solution to this dilemma is to learn the competence, expertise, or assertiveness of metrics. The basic idea is that each metric has a specific sub-domain for which it is most competent (i.e., it consistently produces more accurate classifiers than the ones produced by other metrics). Particularly, we investigate stacking-based meta-learning methods, which use the training data to find the domain of competence of each metric. The meta-classifier describes the domains of competence (or areas of expertise) of each metric, enabling a more sensible use of these metrics so that competence-conscious classifiers can be produced (i.e., a metric is only used to produce classifiers for test instances that belong to its domain of competence). We conducted a systematic evaluation, using different datasets and evaluation measures, of classifiers produced by different metrics. The result is that, while no metric is always superior than all others, the selection of appropriate metrics according to their competence/expertise (i.e., competence-conscious associative classifiers) seems very effective, showing gains that range from 7% to 26% when compared to the baselines (SVMs and an existing ensemble method).

A number of online video social networks, out of which YouTube is the most popular, provides features that allow users to post a video as a response to a discussion topic. These features open opportunities for users to introduce polluted content, or simply pollution, into the system. For instance, spammers may post an unrelated video as response to a popular one aiming at increasing the likelihood of the response being viewed by a larger number of users. Moreover, opportunistic users--promoters--may try to gain visibility to a specific video by posting a large number of (potentially unrelated) responses to boost the rank of the responded video, making it appear in the top lists maintained by the system. Content pollution may jeopardize the trust of users on the system, thus compromising its success in promoting social interactions. In spite of that, the available literature is very limited in providing a deep understanding of this problem.

Abstract In this article we present FLUX‐CiM, a novel method for extracting components (e.g., author names, article titles, venues, page numbers) from bibliographic citations. Our method does not rely on patterns encoding specific delimiters used in a particular citation style. This feature yields a high degree of automation and flexibility, and allows FLUX‐CiM to extract from citations in any given format. Differently from previous methods that are based on models learned from user‐driven training, our method relies on a knowledge base automatically constructed from an existing set of sample metadata records from a given field (e.g., computer science, health sciences, social sciences, etc.). These records are usually available on the Web or other public data repositories. To demonstrate the effectiveness and applicability of our proposed method, we present a series of experiments in which we apply it to extract bibliographic data from citations in articles of different fields. Results of these experiments exhibit precision and recall levels above 94% for all fields, and perfect extraction for the large majority of citations tested. In addition, in a comparison against a state‐of‐the‐art information‐extraction method, ours produced superior results without the training phase required by that method. Finally, we present a strategy for using bibliographic data resulting from the extraction process with FLUX‐CiM to automatically update and expand the knowledge base of a given domain. We show that this strategy can be used to achieve good extraction results even if only a very small initial sample of bibliographic records is available for building the knowledge base.

Due to the increasing amount of information being stored and accessible through the Web, Automatic Document Classification (ADC) has become an important research topic. ADC usually employs a supervised learning strategy, where we first build a classification model using pre-classified documents and then use it to classify unseen documents. One major challenge in building classifiers is dealing with the temporal evolution of the characteristics of the documents and the classes to which they belong. However, most of the current techniques for ADC do not consider this evolution while building and using the models. Previous results show that the performance of classifiers may be affected by three different temporal effects (class distribution, term distribution and class similarity). Further, it is shown that using just portions of the pre-classified documents, which we call contexts, for building the classifiers, result in better performance, as a consequence of the minimization of the aforementioned effects.

Digital Libraries are complex information systems that involve rich sets of digital objects and their respective metadata, along with multiple organizational structures and services (e.g., searching, browsing, and personalization), and are normally built having a target community of users with specific interests. Central to the success of this type of system is the quality of their services and content. In the context of DLs of scientific literature, among the many problems faced to sustain their information quality, two specific ones, related to information consistency, have taken a lot of attention from the research community: name disambiguation and lack of information to access the full-text of cataloged documents. In this paper, we examine these two problems and describe the solutions we have proposed to solve them.

Researchers have been studying and developing teaching materials for information retrieval (IR), such as [3]. Toolkits also have been built that provide hands-on experience to students. For example, IR-Toolbox [4] is an effort to close the gap between the students' understanding of IR concepts and real-life indexing and search systems. Such tools might be good for helping students in non-technical areas such as in the Library and Information Science field to develop their conceptual model of search engines. However, they do not cover emerging topics and skills, such as content-based image retrieval (CBIR) and fusion search. Although there is open source software (such as those in http://www.searchtools.com/tools/tools-opensource.html) that can be used to teach basic and advanced IR topics, they require a student to have high-level technical knowledge and to spend a long time to gain a practical understanding of these topics.

Some applications have to present their results in the form of ranked lists. This is the case of many information retrieval applications, in which documents must be sorted according to their relevance to a given query. This has led the interest of the information retrieval community in methods that automatically learn effective ranking functions. In this paper we propose a novel method which uncovers patterns (or rules) in the training data associating features of the document with its relevance to the query, and then uses the discovered rules to rank documents. To address typical problems that are inherent to the utilization of association rules (such as missing rules and rule explosion), the proposed method generates rules on a demand-driven basis, at query-time. The result is an extremely fast and effective ranking method. We conducted a systematic evaluation of the proposed method using the LETOR benchmark collections. We show that generating rules on a demand-driven basis can boost ranking performance, providing gains ranging from 12 % to 123%, outperforming the state-of-the-art methods that learn to rank, with no need of time-consuming and laborious pre-processing. As a highlight, we also show that additional information, such as query terms, can make the generated rules more discriminative, further improving ranking performance.

Due to the increasing amount of information present on the Web, Automatic Document Classification (ADC) has become an important research topic. ADC usually follows a standard supervised learning strategy, where we first build a model using preclassified documents and then use it to classify new unseen documents. One major challenge for ADC in many scenarios is that the characteristics of the documents and the classes to which they belong may change over time. However, most of the current techniques for ADC are applied without taking into account the temporal evolution of the collection of documents

In this paper we consider the problem of using the block structure of a Web page to improve ranking results when searching for information on Web sites. Given the block structure of the Web pages as input, we propose a method for computing the importance of each block (in the form of block weights) in a Web collection. As we show through experiments, the deployment of our method may allow a significant improvement in the quality of search results. We ran experiments to compare the quality of search results when using our method to the quality obtained when using no structure information. When compared to a ranking method that considered pages as monolithic units, our block-based ranking method led to improvements in the quality of search results in experiments with two sites with heterogeneous structures. Further, our method does not increase the cost of processing queries when compared to the systems using no structural information.

In this paper, we propose a new method to discover collection-adapted ranking functions based on Genetic Programming (GP). Our Combined Component Approach (CCA)is based on the combination of several term-weighting components (i.e.,term frequency, collection frequency, normalization) extracted from well-known ranking functions. In contrast to related work, the GP terminals in our CCA are not based on simple statistical information of a document collection, but on meaningful, effective, and proven components. Experimental results show that our approach was able to outper form standard TF-IDF, BM25 and another GP-based approach in two different collections. CCA obtained improvements in mean average precision up to 40.87% for the TREC-8 collection, and 24.85% for the WBR99 collection (a large Brazilian Web collection), over the baseline functions. The CCA evolution process also was able to reduce the overtraining, commonly found in machine learning methods, especially genetic programming, and to converge faster than the other GP-based approach used for comparison.

We present a novel approach for classifying documents that combines different pieces of evidence (e.g., textual features of documents, links, and citations) transparently, through a data mining technique which generates rules associating these pieces of evidence to predefined classes. These rules can contain any number and mixture of the available evidence and are associated with several quality criteria which can be used in conjunction to choose the "best" rule to be applied at classification time. Our method is able to perform evidence enhancement by link forwarding/backwarding (i.e., navigating among documents related through citation), so that new pieces of link-based evidence are derived when necessary. Furthermore, instead of inducing a single model (or rule set) that is good on average for all predictions, the proposed approach employs a lazy method which delays the inductive process until a document is given for classification, therefore taking advantage of better qualitative evidence coming from the document. We conducted a systematic evaluation of the proposed approach using documents from the ACM Digital Library and from a Brazilian Web directory. Our approach was able to outperform in both collections all classifiers based on the best available evidence in isolation as well as state-of-the-art multi-evidence classifiers. We also evaluated our approach using the standard WebKB collection, where our approach showed gains of 1% in accuracy, being 25 times faster. Further, our approach is extremely efficient in terms of computational performance, showing gains of more than one order of magnitude when compared against other multi-evidence classifiers.

Content-targeted advertising, the task of automatically associating ads to a Web page, constitutes a key Web monetization strategy nowadays. Further, it introduces new challenging technical problems and raises interesting questions. For instance, how to design ranking functions able to satisfy conflicting goals such as selecting advertisements (ads) that are relevant to the users and suitable and profitable to the publishers and advertisers? In this paper we propose a new framework for associating ads with web pages based on Genetic Programming (GP). Our GP method aims at learning functions that select the most appropriate ads, given the contents of a Web page. These ranking functions are designed to optimize overall precision and minimize the number of misplacements. By using a real ad collection and web pages from a newspaper, we obtained a gain over a state-of-the-art baseline method of 61.7% in average precision. Further, by evolving individuals to provide good ranking estimations, GP was able to discover ranking functions that are very effective in placing ads in web pages while avoiding irrelevant ones.

Abstract Traditional text‐based document classifiers tend to perform poorly on the Web. Text in Web documents is usually noisy and often does not contain enough information to determine their topic. However, the Web provides a different source that can be useful to document classification: its hyperlink structure. In this work, the authors evaluate how the link structure of the Web can be used to determine a measure of similarity appropriate for document classification. They experiment with five different similarity measures and determine their adequacy for predicting the topic of a Web page. Tests performed on a Web directory show that link information alone allows classifying documents with an average precision of 86%. Further, when combined with a traditional text‐based classifier, precision increases to values of up to 90%, representing gains that range from 63 to 132% over the use of text‐based classification alone. Because the measures proposed in this article are straightforward to compute, they provide a practical and effective solution for Web classification and related information retrieval tasks. Further, the authors provide an important set of guidelines on how link structure can be used effectively to classify Web documents.

In this paper, we propose a novel framework using Genetic Programming to combine image database descriptors for content-based image retrieval (CBIR). Our framework is validated through several experiments involving two image databases and specific domains, where the images are retrieved based on the shape of their objects.

This paper shows how citation-based information and structural content (e.g., title, abstract) can be combined to improve classification of text documents into predefined categories. We evaluate different measures of similarity -- five derived from the citation information of the collection, and three derived from the structural content -- and determine how they can be fused to improve classification effectiveness. To discover the best fusion framework, we apply Genetic Programming (GP) techniques. Our experiments with the ACM Computing Classification Scheme, using documents from the ACM Digital Library, indicate that GP can discover similarity functions superior to those based solely on a single type of evidence. Effectiveness of the similarity functions discovered through simple majority voting is better than that of content-based as well as combination-based Support Vector Machine classifiers. Experiments also were conducted to compare the performance between GP techniques and other fusion techniques such as Genetic Algorithms (GA) and linear fusion. Empirical results show that GP was able to discover better similarity functions than GA or other fusion techniques.

This paper shows how different measures of similarity derived from the citation information and the structural content (e.g., title, abstract) of the collection can be fused to improve classification effectiveness. To discover the best fusion framework, we apply Genetic Programming (GP) techniques. Our experiments with the ACM Computing Classification Scheme, using documents from the ACM Digital Library, indicate that GP can discover similarity functions superior to those based solely on a single type of evidence. Effectiveness of the similarity functions discovered through simple majority voting is better than that of content-based as well as combination-based Support Vector Machine classifiers. Experiments also were conducted to compare the performance between GP techniques and other fusion techniques such as Genetic Algorithms (GA) and linear fusion. Empirical results show that GP was able to discover better similarity functions than other fusion techniques.

This paper discusses how citation-based information and structural content (e.g., title, abstract) can be combined to improve classification of text documents into predefined categories. We evaluate different measures of similarity derived from the citation structure and the structural content of the collection, and determine how they can be fused to improve classification effectiveness. To discover the best fusion framework, we apply Genetic Programming (GP) techniques. Our empirical experiments using documents from the ACM Digital Library and the ACM Computing Classification System show that we can discover similarity functions that work better than using evidence in isolation and whose combined performance through a simple majority voting is comparable to that of Support Vector Machine classifiers.

Digital libraries (DLs) are complex information systems and therefore demand formal foundations lest development efforts diverge and interoperability suffers. In this article, we propose the fundamental abstractions of Streams, Structures, Spaces, Scenarios, and Societies (5S), which allow us to define digital libraries rigorously and usefully. Streams are sequences of arbitrary items used to describe both static and dynamic (e.g., video) content. Structures can be viewed as labeled directed graphs, which impose organization. Spaces are sets with operations on those sets that obey certain constraints. Scenarios consist of sequences of events or actions that modify states of a computation in order to accomplish a functional requirement. Societies are sets of entities and activities and the relationships among them. Together these abstractions provide a formal foundation to define, relate, and unify concepts---among others, of digital objects, metadata, collections, and services---required to formalize and elucidate "digital libraries". The applicability, versatility, and unifying power of the 5S model are demonstrated through its use in three distinct applications: building and interpretation of a DL taxonomy, informal and formal analysis of case studies of digital libraries (NDLTD and OAI), and utilization as a formal basis for a DL description language.

This paper studies how link information can be used to improve classification results for Web collections. We evaluate four different measures of subject similarity, derived from the Web link structure, and determine how accurate they are in predicting document categories. Using a Bayesian network model, we combine these measures with the results obtained by traditional content-based classifiers. Experiments on a Web directory show that best results are achieved when links from pages outside the directory are considered. Link information alone is able to obtain gains of up to 46 points in F1, when compared to a traditional content-based classifier. The combination with content-based methods can further improve the results, but too much noise may be introduced, since the text of Web pages is a much less reliable source of information. This work provides an important insight on which measures derived from links are more appropriate to compare Web documents and how these measures can be combined with content-based algorithms to improve the effectiveness of Web classification.

The Web contains a huge volume of information, almost all unstructured and, therefore, difficult to manage. In Digital Libraries, however, information is explicitly organized, described, and managed. In this paper, we propose an architecture that allows the construction of digital libraries from the Web, using standard protocols and archival technologies, and incorporating powerful digital library and data extraction tools, thus benefiting from the breadth of the Web contents, but supporting services and organization available in digital libraries. The proposed architecture was applied to the Networked Digital Library of Theses and Dissertations, providing an important first step toward rapid construction of large DLs from the Web, as well as a large-scale solution for interoperability between independent digital libraries.

In this demonstration, we present interoperable and personalized search services for the Networked Digital Library of Theses and Dissertations (NDLTD). Using standard protocols and software, including those specified by the Open Archives Initiative (OAI), distributed sites can share metadata easily. On top of these harvesting protocols, we implement a union collection of theses managed by the MARIAN digital library system. Our demonstration covers aspects of NDLTD, OAI, and MARIAN.