Abstract News agencies report news from different viewpoints and with different writing styles and these differences often ap pear in their article descriptions. We propose a method to extract characteristic descriptions on certain entities (per sons, locations, organizations, etc.) in news agency articles. For a given entity, a description is one tuple (called an SVO tuple) consisting of that entity and... News agencies report news from different viewpoints and with different writing styles and these differences often ap pear in their article descriptions. We propose a method to extract characteristic descriptions on certain entities (per sons, locations, organizations, etc.) in news agency articles. For a given entity, a description is one tuple (called an SVO tuple) consisting of that entity and other words or phrases appearing in the same sentence on the basis of their SVO (Subject (S), Verb (V) and Object (O)) roles. We extract characteristic descriptions of entities by computing the fre quency and inverse agency frequency of each description. By arranging characteristic descriptions chronologically, we were able to ascertain the left-or right-leanings of news agencies and how those leanings varied with the passage of time. Read More Read Less

Abstract With the dramatic increasing of electronic Arabic content, the text compression techniques will play a major role in several domains and applications such as search engines, data archiving, searching and retrieval from huge databases. Mainly the combination of compression and indexing techniques allows the interesting possibility to work directly on the compressed textual fi les or databases, which results saving... With the dramatic increasing of electronic Arabic content, the text compression techniques will play a major role in several domains and applications such as search engines, data archiving, searching and retrieval from huge databases. Mainly the combination of compression and indexing techniques allows the interesting possibility to work directly on the compressed textual fi les or databases, which results saving time and resources. The existing compression techniques and tools are generic and do not consider the specifi c characteristics of the Arabic language such as its derivative nature. Mainly compression techniques should be based on the morphology characteristics of the Arabic language, its grammatical characteristics, the texts subject, and their statistical characteristics. The paper surveys the state of the art of the Arabic texts compression techniques and tools and identifi es some research tracks that should be explored in future. It presents also some dedicated Arabic text compression algorithms which save more physical space and speed up the data retrieval text fi les by searching in their compressed form. Read More Read Less

Abstract During web search, confusion can happen due to homonym when users use non-unique values as a search term of an entity. Especially, when parts of names of an entity were used as its identifi er, we call a mixed entity resolution problem whose goal is to clear out the erroneous entities. For example, if only last name is used as... During web search, confusion can happen due to homonym when users use non-unique values as a search term of an entity. Especially, when parts of names of an entity were used as its identifi er, we call a mixed entity resolution problem whose goal is to clear out the erroneous entities. For example, if only last name is used as an identifi er, we cannot distinguish “Vannessa Bush” from “George Bush.” Mixed entity resolution problem is common among Web pages data. In this paper, to resolve aforementioned mixed entities on the Web, we propose a prototypical system which includes a web service based interface, unsupervised clustering scheme, and cluster ranking algorithms. In particular, since the correct number of clusters is often unknown, we study a state-of-the-art unsupervised clustering solution based on propagation of pairwise similarities of entities. Experimental results show that our approach outperforms main competing solution. Read More Read Less

Abstract Digital libraries in their current form are bounded by their ineffcient webpage-based user interface paradigm, and even the most knowledgeable researchers can get lost in the large amount of published material available. A paper and its immediate references are displayed on a single webpage. Unfortunately, it is not readily apparent where the paper belongs in the greater context of a... Digital libraries in their current form are bounded by their ineffcient webpage-based user interface paradigm, and even the most knowledgeable researchers can get lost in the large amount of published material available. A paper and its immediate references are displayed on a single webpage. Unfortunately, it is not readily apparent where the paper belongs in the greater context of a research fi eld. Our goal was to develop, test, and iteratively improve the effectiveness of a web-based application, PaperCube, designed to augment and enhance a researcher’s interaction with digital libraries through the use of interactive visualizations of bibliographic meta data. The visualizations set out to show different levels of detail and aspects of a paper’s citation network. A primary concern was to ensure that while switching visualizations, the viewed paper or author remained focused at all times. PaperCube was validated through a user study that showed that it helped users who are trying to explore a fi eld of research. In particular, PaperCube was shown to reduce the “cognitive load” put on a researcher and aid the “discoverability” of new research material. Read More Read Less

Abstract The global diffusion of the Internet has enabled the distribution of informative content through dynamic media such as RSS feeds and video blogs. At the same time, the decreasing cost of electronic devices has increased the pervasive availability of the same informative content in the form of digital audiovisual data. This article presents a system for the large-scale unsupervised acquisition,... The global diffusion of the Internet has enabled the distribution of informative content through dynamic media such as RSS feeds and video blogs. At the same time, the decreasing cost of electronic devices has increased the pervasive availability of the same informative content in the form of digital audiovisual data. This article presents a system for the large-scale unsupervised acquisition, segmentation and indexing of TV newscasts. In particular, it discusses the principles and performance of the parts of the system dedicated to the detection and segmentation of programmes from the acquired stream. In addition to the core technology, we also introduce and discuss a novel method for assessing the results of story boundaries segmentation algorithms, based on a user-validated measurement. Due to the heterogeneity of current news distribution channels, a further innovative aspect of this article is the description of a framework for multimodal information aggregation. The core of this framework is a cross-modal clustering process for which a novel, asymmetric similarity measure is provided. The implemented prototype uses online news articles and TV news programmes as information sources, and provides a multimodal service integrating both contributions. Experimental evaluation of the system proves the effectiveness of the method in the studied case. Read More Read Less

Abstract Choosing a restaurant is one of the most frequent decisions faced in modern daily life; however, it is diffi cult for consumers to choose between food/restaurant by reading large amounts of reviews. This study attempts to generate cuisine hotspot maps through blog content mining to help consumers make restaurant decisions by specialties. The main obstacle in doing this involves recognizing... Choosing a restaurant is one of the most frequent decisions faced in modern daily life; however, it is diffi cult for consumers to choose between food/restaurant by reading large amounts of reviews. This study attempts to generate cuisine hotspot maps through blog content mining to help consumers make restaurant decisions by specialties. The main obstacle in doing this involves recognizing and extracting restaurants and essentialrestaurant information (i.e., restaurant dishes) in unstructured content. In contrast to traditional Named Entity Recognition (NER) targets, dish name is a promising target that received little attention in previous studies. This study develops methods for recognizing and extracting restaurant names and dish names from review postsin the blogosphere and achieves satisfactory performance. Based on the method, we processed more than 12,000 Chinese blog posts and generated a cuisine hotspot map. The map shows the most popular dishes of restaurants in a map-view to help consumers make restaurant decisions. A prototype of cuisine hotspot map, named CuisineGuide, is implemented and available as an iPhone application. Read More Read Less