Marco Cristo

Learning to advertise

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.

Combining link-based and content-based methods for web document classification

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.

Automatic quality assessment of content created collaboratively by web communities: a case study of wikipedia

The old dream of a universal repository containing all the human knowledge and culture is becoming possible through the Internet and the Web. Moreover, this is happening with the direct collaborative, participation of people. Wikipedia is a great example. It is an enormous repository of information with free access and edition, created by the community in a collaborative manner. However, this large amount of information, made available democratically and virtually without any control, raises questions about its relative quality. In this work we explore a significant number of quality indicators, some of them proposed by us and used here for the first time, and study their capability to assess the quality of Wikipedia articles. Furthermore, we explore machine learning techniques to combine these quality indicators into one single assessment judgment. Through experiments, we show that the most important quality indicators are the easiest ones to extract, namely, textual features related to length, structure and style. We were also able to determine which indicators did not contribute significantly to the quality assessment. These were, coincidentally, the most complex features, such as those based on link analysis. Finally, we compare our combination method with state-of-the-art solution and show significant improvements in terms of effective quality prediction.

A combined component approach for finding collection-adapted ranking functions based on genetic programming

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.

Exploiting user feedback to learn to rank answers in q&a forums: a case study with stack overflow

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.

A comparative study of citations and links in document classification

It is well known that links are an important source of information when dealing with Web collections. However, the question remains on whether the same techniques that are used on the Web can be applied to collections of documents containing citations between scientific papers. In this work we present a comparative study of digital library citations and Web links, in the context of automatic text classification. We show that there are in fact differences between citations and links in this context. For the comparison, we run a series of experiments using a digital library of computer science papers and a Web directory. In our reference collections, measures based on co-citation tend to perform better for pages in the Web directory, with gains up to 37% over text based classifiers, while measures based on bibliographic coupling perform better in a digital library. We also propose a simple and effective way of combining a traditional text based classifier with a citation-link based classifier. This combination is based on the notion of classifier reliability and presented gains of up to 14% in micro-averaged F1 in the Web collection. However, no significant gain was obtained in the digital library. Finally, a user study was performed to further investigate the causes for these results. We discovered that misclassifications by the citation-link based classifiers are in fact difficult cases, hard to classify even for humans

Automatic Assessment of Document Quality in Web Collaborative Digital Libraries

The old dream of a universal repository containing all of human knowledge and culture is becoming possible through the Internet and the Web. Moreover, this is happening with the direct collaborative participation of people. Wikipedia is a great example. It is an enormous repository of information with free access and open edition, created by the community in a collaborative manner. However, this large amount of information, made available democratically and virtually without any control, raises questions about its quality. In this work, we explore a significant number of quality indicators and study their capability to assess the quality of articles from three Web collaborative digital libraries. Furthermore, we explore machine learning techniques to combine these quality indicators into one single assessment. Through experiments, we show that the most important quality indicators are those which are also the easiest to extract, namely, the textual features related to the structure of the article. Moreover, to the best of our knowledge, this work is the first that shows an empirical comparison between Web collaborative digital libraries regarding the task of assessing article quality.

Evidence of quality of textual features on the web 2.0

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. In this context, this work aims at assessing the relative quality of distinct textual features available on the Web 2.0. Towards this goal, we analyzed four features (title, tags, description and comments) in four popular applications (CiteULike, Last.FM, Yahoo! Video, and Youtube). Firstly, we characterized data from these applications in order to extract evidence of quality of each feature with respect to usage, amount of content, descriptive and discriminative power as well as of content diversity across features. Afterwards, a series of classification experiments were conducted as a case study for quality evaluation. Characterization and classification results indicate that: 1) when considered separately, tags is the most promising feature, achieving the best classification results, although its absence in a non-negligible fraction of objects may affect its potential use; and 2) each feature may bring different pieces of information, and combining their contents can improve classification.

Bayesian belief networks for IR

We review the application of Bayesian belief networks to several information retrieval problems, showing that they provide an effective and flexible framework for modeling distinct sources of evidence in support of a ranking. To illustrate, we explain how Bayesian networks can be used to represent the classic vector space model and demonstrate how this basic representation can be extended to naturally incorporate new evidence from distinct information sources. These models have been shown useful in several text collections, where the combination of evidential information derived from past queries, thesauri, and the link structure of Web pages has led to significant improvements in retrieval performance.

An incremental technique for real-time bioacoustic signal segmentation

An incremental transformation of ZCR and energy without using temporal windows.With our method is possible to save memory and transmission costs.Solution to process large amounts of data by resource-constrained devices as WSN. A bioacoustical animal recognition system is composed of two parts: (1) the segmenter, responsible for detecting syllables (animal vocalization) in the audio; and (2) the classifier, which determines the species/animal whose the syllables belong to. In this work, we first present a novel technique for automatic segmentation of anuran calls in real time; then, we present a method to assess the performance of the whole system. The proposed segmentation method performs an unsupervised binary classification of time series (audio) that incrementally computes two exponentially-weighted features (Energy and Zero Crossing Rate). In our proposal, classical sliding temporal windows are replaced with counters that give higher weights to new data, allowing us to distinguish between a syllable and ambient noise (considered as silences). Compared to sliding-window approaches, the associated memory cost of our proposal is lower, and processing speed is higher. Our evaluation of the segmentation component considers three metrics: (1) the Matthews Correlation Coefficient for point-to-point comparison; (2) the WinPR to quantify the precision of boundaries; and (3) the AEER for event-to-event counting. The experiments were carried out in a dataset with 896 syllables of seven different species of anurans. To evaluate the whole system, we derived four equations that helps understand the impact that the precision and recall of the segmentation component has on the classification task. Finally, our experiments show a segmentation/recognition improvement of 37%, while reducing memory and data communication. Therefore, results suggest that our proposal is suitable for resource-constrained systems, such as Wireless Sensor Networks (WSNs).