Sabine Loudcher

A Joint Model for Topic-Sentiment Evolution over Time

Most existing topic models focus either on extracting static topic-sentiment conjunctions or topic-wise evolution over time leaving out topic-sentiment dynamics and missing the opportunity to provide a more in-depth analysis of textual data. In this paper, we propose an LDA-based topic model for analyzing topic-sentiment evolution over time by modeling time jointly with topics and sentiments. We derive inference algorithm based on Gibbs Sampling process. Finally, we present results on reviews and news datasets showing interpretable trends and strong correlation with ground truth in particular for topic-sentiment evolution over time.

Warehousing complex data from the web

Data warehousing and Online Analytical Processing (OLAP) technologies are now moving onto handling complex data that mostly originate from the web. However, integrating such data into a decision-support process requires their representation in a form processable by OLAP and/or data mining techniques. We present in this paper a complex data warehousing methodology that exploits eXtensible Markup Language (XML) as a pivot language. Our approach includes the integration of complex data in an ODS, in the form of XML documents; their dimensional modelling and storage in an XML data warehouse; and their analysis with combined OLAP and data mining techniques. We also address the crucial issue of performance in XML warehouses.

A joint model for topic-sentiment modeling from text

Traditional topic models, like LDA and PLSA, have been efficiently extended to capture further aspects of text in addition to the latent topics (e.g., time evolution, sentiment etc.). In this paper, we discuss the issue of joint topic-sentiment modeling. We propose a novel topic model for topic-specific sentiment modeling from text and we derive an inference algorithm based on the Gibbs sampling process. We also propose a method for automatically setting the model parameters. The experiments performed on two review datasets show that our model outperforms other state-of-the-art models, in particular for sentiment prediction at the topic level.

Combining OLAP and information networks for bibliographic data analysis: a survey

In the context of scientometrics and bibliometrics, several research fields are dealing with bibliographic data. In this paper, we will explore how the combination of online analytical processing (OLAP) analysis and information networks could be an interesting issue. In Business Intelligence, OLAP is a technology supported by data warehousing systems. It provides tools for analyzing data according to multiple dimensions and multiple hierarchical levels. At the same time, several information networks (co-authors network, citations network, institutions network, etc.) can be built based on bibliographic databases. Originally, OLAP was introduced to analyze structured data. However, in this paper, we wonder if, by combining OLAP and information networks, we can provide a new way of analyzing bibliographic data. OLAP should be able to handle information networks and be also useful for monitoring, browsing and analyzing the content and the structure of bibliographic networks. The goal of this survey paper is to review previous work on OLAP and information networks dealing with bibliographic data. We also propose a comparison between traditional OLAP and OLAP on information networks and discuss the challenges OLAP faces regarding bibliographic networks.

OLAP on Information Networks: A New Framework for Dealing with Bibliographic Data

In the context of decision making, data warehouses support OLAP technology and they have been very useful for efficient analysis onto structured data. For several years, OLAP is also used to analyze and visualize more complex data. Now, many data sets of interest can be described as a linked collection of interrelated objects. They could be represented as heterogeneous information networks, in which there are multiple object and link types. In this paper, we are focusing on bibliographic data. This type of data constitutes a rich source that is the starting point of research on bibliometrics, scientometrics domains. In this context, we discuss the interest of combining information networks, OLAP and data mining technologies. We propose a framework to materialize this combination and discuss the main challenges to build this framework. The basic idea is to be able to analyze various networks built from the bibliographic data representing different points of view (authors networks, citations networks...) and their dynamic.

OLAP on Complex Data: Visualization Operator Based on Correspondence Analysis

Data warehouses and Online Analysis Processing (OLAP) have acknowledged and efficient solutions for helping in the decision-making process. Through OLAP operators, online analysis enables the decision-maker to navigate and view data represented in a multi-dimensional manner. But when the data or objects to be analyzed are complex, it is necessary to redefine and enhance the abilities of the OLAP. In this paper, we suggest combining OLAP and data mining in order to create a new visualization operator for complex data or objects. This operator uses the correspondence analysis method and we call it VOCoDa (Visualization Operator for Complex Data).

Graphs enriched by cubes for OLAP on bibliographic networks

With the recent growth of bibliographic data, many research fields work on defining new techniques for their analysis. In this context, data could be represented as heterogeneous networks. In order to analyse information networks in a multidimensional way, online analytical processing OLAP may be a relevant solution but it must be adapted for networked data by considering nodes and edges. A first approach that has been proposed in the literature consists in building cubes of graphs. In a different and complementary way, our proposal consists in enriching graphs with cubes. Indeed, the nodes or/and edges of the considered network are described by a cube. It allows interesting analyses for the user who can navigate within a graph enriched by cubes according to different granularity levels, with dedicated operators. We implemented our approach and performed an experimental study on a real dataset to show the interest of our proposal.

GOTA - Using the Google Similarity Distance for OLAP Textual Aggregation

With the tremendous growth of unstructured data in the Business Intelligence, there is a need for incorporating textual data into data warehouses, to provide an appropriate multidimensional analysis (OLAP) and develop new approaches that take into account the textual content of data. This will provide textual measures to users who wish to analyse documents online. In this paper, we propose a new aggregation function for textual data in an OLAP context. For aggregating keywords, our contribution is to use a data mining technique, such as kmeans, but with a distance based on the Google similarity distance. Thus our approach considers the semantic similarity of keywords for their aggregation. The performance of our approach is analyzed and compared to another method using the k-bisecting clustering algorithm and based on the Jensen-Shannon divergence for the probability distributions. The experimental study shows that our approach achieves better performances in terms of recall, precision,F-measure complexity and runtime.

Efficiently mining frequent itemsets applied for textual aggregation

Text mining approaches are commonly used to discover relevant information and relationships in huge amounts of text data. The term data mining refers to methods for analyzing data with the objective of finding patterns that aggregate the main properties of the data. The merger between the data mining approaches and on-line analytical processing (OLAP) tools allows us to refine techniques used in textual aggregation. In this paper, we propose a novel aggregation function for textual data based on the discovery of frequent closed patterns in a generated documents/keywords matrix. Our contribution aims at using a data mining technique, mainly a closed pattern mining algorithm, to aggregate keywords. An experimental study on a real corpus of more than 700 scientific papers collected on Microsoft Academic Search shows that the proposed algorithm largely outperforms four state-of-the-art textual aggregation methods in terms of recall, precision, F-measure and runtime.

Fusion Techniques for Named Entity Recognition and Word Sense Induction and Disambiguation

In this paper we explore the use of well-known multimodal fusion techniques to solve two prominent Natural Language Processing tasks. Specifically, we focus on solving Named Entity Recognition and Word Sense Induction and Disambiguation by applying feature-combination methods that have already shown their efficiency in the multimedia analysis domain. We present a series of experiments employing fusion techniques in order to combine textual linguistic features. Our intuition is that by combining different types of features we may find semantic relatedness among words at different levels and thus, the combination (and recombination) of these levels may yield gains in terms of metrics’ performance. To our knowledge, employing these techniques has not been studied for the tasks we address in this paper. We test the proposed fusion techniques on three datasets for named entity recognition and one for word sense disambiguation and induction. Our results show that the combination of textual features indeed improves the performance compared to single feature representation and the trivial feature concatenation.