Jean-Yves Delort’s research while affiliated with Google Inc. and other places

This paper describes the generation of temporally anchored infobox attribute data from the Wikipedia history of revisions. By mining (attribute, value) pairs from the revision history of the English Wikipedia we are able to collect a comprehensive knowledge base that contains data on how attributes change over time. When dealing with the Wikipedia edit history, vandalic and erroneous edits are a concern for data quality. We present a study of vandalism identification in Wikipedia edits that uses only features from the infoboxes, and show that we can obtain, on this dataset, an accuracy comparable to a state-of-the-art vandalism identification method that is based on the whole article. Finally, we discuss different characteristics of the extracted dataset, which we make available for further study.

We describe the use of a hierarchical topic model for automatically identifying syntactic and lexical patterns that explicitly state ontological relations. We leverage distant supervision using relations from the knowledge base FreeBase, but do not require any manual heuristic nor manual seed list selections. Results show that the learned patterns can be used to extract new relations with good precision.

Update summarization is a new challenge in multi-document summarization focusing on summarizing a set of recent documents relatively to another set of earlier documents. We present an unsupervised probabilistic approach to model novelty in a document collection and apply it to the generation of update summaries. The new model, called Dualsum, results in the second or third position in terms of the ROUGE metrics when tuned for previous TAC competitions and tested on TAC-2011, being statistically indistinguishable from the winning system. A manual evaluation of the generated summaries shows state-of-the art results for Dualsum with respect to focus, coherence and overall responsiveness.

Online discussion sites are plagued with various types of unwanted content, such as spam and obscene or malicious messages. Prevention and detection-based techniques have been proposed to filter inappropriate content from online discussion sites. But, even though prevention techniques have been widely adopted, detection of inappropriate content remains mostly a manual task. Existing detection techniques, which are divided into rule-based and statistical techniques, suffer from various limitations. Rule-based techniques usually consist of manually crafted rules or blacklists of key words. Both are time-consuming to create and tend to generate many false-positives and false-negatives. Statistical techniques typically use corpora of labeled examples to train a classifier to tell "good" and "bad" messages apart. Although statistical techniques are generally more robust than rule-based techniques, they are difficult to deploy because of the prohibitive cost of manually labeling examples. In this paper we describe a novel classification technique to train a classifier from a partially labeled corpus and use it to moderate inappropriate content on online discussion sites. Partially labeled corpora are much easier to produce than completely labeled corpora, as they are made up only with unlabeled examples and examples labeled with a single class (e.g., "bad"). We implemented and tested this technique on a corpus of messages posted on a stock message board and compared it with two baseline techniques. Results show that our method outperforms the two baselines and that it can be used to significantly reduce the number of messages that need to be reviewed by human moderators.

This paper presents a technique for visualizing large spatial data sets in Web Mapping Systems (WMS). The technique creates a hierarchical clustering tree, which is subsequently used to extract clusters that can be displayed at a given scale without cluttering the map. Voronoi polygons are used as aggregation symbols to represent the clusters. This technique retains hierarchical relationships between data items at different scales. In addition, aggregation symbols do not overlap, and their sizes and the number of points that they cover is controlled by the same parameter. A prototype has been implemented and tested showing the effectiveness of the method for visualizing large data sets in WMS.

This paper addresses the problem of reducing cluttering in interactive maps. It presents a new technique for visualizing large spatial datasets using hierarchical aggregation. The technique creates a hierarchical clustering tree, which is subsequently used to extract clusters that can be displayed at a given scale without cluttering the map. Voronoi polygons are used as aggregation symbols to represent the clusters. This technique retains hierarchical relationships between data items at different scales. In addition, aggregation symbols do not overlap, and their sizes and the number of points that they cover is controlled by the same parameter. The scalability analysis shows that the method can effectively be used with datasets of up to 1000 items.

Financial surveillance technology alerts analysts to suspicious trading events. Our aim is to identify explainable false positives (e.g., caused by price-sensitive information in company news) and explainable true positives (e.g., caused by ramping in forums) by aligning these alerts with publicly available information. Our system aligns 99% of alerts, which will speed the analysts' task by helping them to eliminate false positives and gather evidence for true positives more rapidly. © Institute for Computer Sciences, Social-Informatics and Telecommunications Engineering 2010.

Internet message boards are often used to spread information in order to manipulate financial markets. Although this hypothesis is supported by many cases reported in the literature and in the media, the real impact of manipulation in online forums on financial markets remains an open question. This work analyses the effect of manipulation in internet stock message boards on financial markets. We employ a unique corpus of moderated messages to investigate market manipulation. Internet message boards administrators use the process of moderation to restrict market manipulation. From the data we find that manual supervision of stock message boards by moderators does not effectively protect users against manipulation. Furthermore, by focusing on messages that have been moderated as manipulative due to ramping we show that ramping is positively related to market returns, volatility and volume. We also demonstrate that stocks with higher turnover, lower price level, lower market capitalization and higher volatility are more common targets of ramping.

Salience is an important characteristic of information influencing users' cognitive and emotional states. For example, salient parts of a document are those that readers will find moving or provoking. This article analyzes the main characteristics of salience and the different meanings of the concept in information retrieval and linguistics. It also presents a generic approach for identifying linguistically salient segments in a text using readers' textual feedback. The method, supports any kind of text and. textual feedback. We evaluated the effectiveness of the method with a. corpus of blog posts and readers' comments. Our preliminary experiments show that the method has promising results with an fscore of 0.65. The method could also be used on 90% of commented posts which proves that it can be used on a large scale.

Le Web compte de plus en plus d'hypercartes qui permettent de visualiser et d'explorer des informations géographiques graphiquement représentées sur une carte. Dans les interfaces actuelles des hypercartes, les utilisateurs ont souvent du mal à naviguer et ils expérimentent souvent un sentiment de "perte dans l'hyperespace". Cet article analyse les principaux facteurs de perte dans l'hyperespace et présente une interface adaptée à la navigation dans les hypercartes.