Nadjia Benblidia

CXT-cube: contextual text cube model and aggregation operator for text OLAP

Traditional data warehousing technologies and On-Line Analytical Processing (OLAP) are unable to analyze textual data. Moreover, as OLAP queries of a decision-maker are generally related to a context, contextual information must be taken into account during the exploitation of data warehouses. Thus, we propose a contextual text cube model denoted CXT-Cube which considers several contextual factors during the OLAP analysis in order to better consider the contextual information associated with textual data. CXT-Cube is characterized by several contextual dimensions, each one related to a contextual factor. In addition, we extend our aggregation OLAP operator for textual data ORank (OLAP-Rank) to consider all the contextual factors defined in our CXT-Cube model. To validate our model, we perform an experimental study and the preliminary results show the importance of our approach for integrating textual data into a data warehouse and improving the decision-making.

Community Extraction Based on Topic-Driven-Model for Clustering Users Tweets

Twitter has become a significant means by which people communicate with the world and describe their current activities, opinions and status in short text snippets. Tweets can be analyzed automatically in order to derive much potential information such as, interesting topics, social influence, user’s communities, etc. Community extraction within social networks has been a focus of recent work in several areas. Different from the most community discovery methods focused on the relations between users, we aim to derive user’s communities based on common topics from user’s tweets. For instance, if two users always talk about politic in their tweets, thus they can be grouped in the same community which is related to politic topic. To achieve this goal, we propose a new approach called CETD: Community Extraction based on Topic-Driven-Model. This approach combines our proposed model used to detect topics of the user’s tweets based on a semantic taxonomy together with a community extraction method based on the hierarchical clustering technique. Our experimentation on the proposed approach shows the relevant of the users communities extracted based on their common topics and domains.

Hybrid image representation methods for automatic image annotation: A survey

In most automatic image annotation systems, images are represented with low level features using either global methods or local methods. In global methods, the entire image is used as a unit. Local methods divide images into blocks where fixed-size sub-image blocks are adopted as sub-units; or into regions by using segmented regions as sub-units in images. In contrast to typical automatic image annotation methods that use either global or local features exclusively, several recent methods have considered incorporating the two kinds of information, and believe that the combination of the two levels of features is beneficial in annotating images. In this paper, we provide a survey on automatic image annotation techniques according to one aspect: feature extraction, and, in order to complement existing surveys in literature, we focus on the emerging image annotation methods: hybrid methods that combine both global and local features for image representation.

Multi-scale salient object detection using graph ranking and global-local saliency refinement

We propose an algorithm for salient object detection (SOD) based on multi-scale graph ranking and iterative local-global object refinement. Starting from a set of multi-scale image decompositions using superpixels, we propose an objective function which is optimized on a multi-layer graph structure to diffuse saliency from image borders to salient objects. This step aims at roughly estimating the location and extent of salient objects in the image. We then enhance the object saliency through an iterative process employing random forests and local boundary refinement using color, texture and edge information. We also use a feature weighting scheme to ensure optimal object/background discrimination. Our algorithm yields very accurate saliency maps for SOD while maintaining a reasonable computational time. Experiments on several standard datasets have shown that our approach outperforms several recent methods dealing with SOD. HighlightsSalient object detection using multi-layered graph ranking.Local-global saliency refinement using objectness-measure and random forests.Region and boundary information for salient object detection.Feature relevance for object/background discrimination.

Contextualized Text OLAP Based on Information Retrieval

Current data warehousing and On-Line Analytical Processing (OLAP) systems are not yet particularly appropriate for textual data analysis. It is therefore crucial to develop a new data model and an OLAP system to provide the necessary analyses for textual data. To achieve this objective, this paper proposes a new approach based on information retrieval (IR) techniques. Moreover, several contextual factors may significantly affect the information relevant to a decision-maker. Thus, the paper proposes to consider contextual factors in an OLAP system to provide relevant results. It provides a generalized approach for Text OLAP analysis which consists of two parts: The first one is a context-based text cube model, denoted CXT-Cube. It is characterized by several contextual dimensions. Hence, during the OLAP analysis process, CXT-Cube exploits the contextual information in order to better consider the semantics of textual data. Besides, the work associates to CXT-Cube a new text analysis measure based on an OLAP-adapted vector space model and a relevance propagation technique. The second part is an OLAP aggregation operator called ORank (OLAP-Rank) which allows to aggregate textual data in an OLAP environment while considering relevant contextual factors. To consider the user context, this paper proposes a query expansion method based on a decision-maker profile. Based on IR metrics, it evaluates the proposed aggregation operator in different cases using several data analysis queries. The evaluation shows that the precision of the system is significantly better than that of a Text OLAP system based on classical IR. This is due to the consideration of the contextual factors.

Combining Context and Content for Automatic Image Annotation on Mobile Phones

Nowadays, the explosive use of mobile phones leads to generate a large number of personal photos. The requirements of effective image retrieval becomes evident. In this paper, we overview image retrieval techniques with special emphasis on different works that consider automatic image annotation as solution to retrieve images in mobile environment. In most automatic image annotation systems, the high-level concepts are generated based, exclusively, on visual analysis or context analysis of the image. Emerging approaches consider that the fusion of content and context analysis is beneficial for automatic image annotation. We focus in this paper on the type of information, visual and contextual, used to generate semantic that describes the image and the new methods that combine both content and context for automatic image annotation.

Event recognition in photo albums using probabilistic graphical model and feature relevance

The exponential use of digital cameras has raised a new problem: how to store/retrieve images/albums in very large photo databases that correspond to special events. In this paper, we propose a new probabilistic graphical model (PGM) to recognize events in photo albums stored by users. The PGM combines high-level image features consisting of scenes and objects detected in images. To consider the discriminative power of features, our model integrates the object/scene relevance for more precise prediction of semantic events in photo albums. Experimental results carried out on the challenging PEC dataset with 807 photo albums are presented.

Event recognition in photo albums using probabilistic graphical models and feature relevance

Event recognition in image albums using high level object/scene features.Convolutional neural networks for scene and object detection in images.Probabilistic graphical models and Bayesian inference for event recognition.Integration of feature relevance for more efficient event recognition. This paper proposes a method for event recognition in photo albums which aims at predicting the event categories of groups of photos. We propose a probabilistic graphical model (PGM) for event prediction based on high-level visual features consisting of objects and scenes, which are extracted directly from images. For better discrimination between different event categories, we develop a scheme to integrate feature relevance in our model which yields a more powerful inference when album images exhibit a large number of objects and scenes. It allows also to mitigate the influence of non-informative images usually contained in the albums. The performance of the proposed method is validated using extensive experiments on the recently-proposed PEC dataset containing over 61 000 images. Our method obtained the highest accuracy which outperforms previous work.

TLabel: A New OLAP Aggregation Operator in Text Cubes

Data Warehousing technologies and On-Line Analytical Processing OLAP feature a wide range of techniques for the analysis of structured data. However, these techniques are inadequate when it comes to analyzing textual data. Indeed, classical aggregation operators have earned their spurs in the online analysis of numerical data, but are unsuitable for the analysis of textual data. To alleviate this shortcoming, on-line analytical processing in text cubes requires new analysis operators adapted to textual data. In this paper, the authors propose a new aggregation operator named Text Label TLabel, based on text categorization. Their operator aggregates textual data in several classes of documents. Each class is associated with a label that represents the semantic content of the textual data of the class. TLabel is founded on a tailoring of text mining techniques to OLAP. To validate their operator, the authors perform an experimental study and the preliminary results show the interest of their approach for Text OLAP.

Latent semantic analysis-based image auto annotation

The image retrieval is a particular case of information retrieval. It adds more complex mechanisms to relevance image retrieval: visual content analysis and/or additional textual content. The image auto annotation is a technique that associates text to image, and permits to retrieve image documents as textual documents, thus as in information retrieval. The image auto annotation is then an effective technology for improving the image retrieval. In this work, we propose the AnnotB-LSA algorithm in its first version for the image auto-annotation. The integration of the LSA model permits to extract the latent semantic relations in the textual describers and to minimize the ambiguousness (polysemy, synonymy) between the annotations of images.