Anilu Franco-Arcega

I, Me, Mine: The Role of Personal Phrases in Author Profiling

The Author Profiling (AP) task aims to distinguish between groups of authors labeled by a common demographic characteristic such as gender or age by studying the language usage. In this work we studied the role of personal phrases (i.e., sentences containing first person pronouns) for the AP task. We support the idea that people better expose their personal interests and writing style when they talk about themselves and, consequently, that words near to a personal pronoun reveal valuable information for the classification of authors. The evaluation using different social media data showed that phrases containing singular first person pronouns are highly valuable for predicting the age and gender of users. Considering only these phrases we obtained reductions of up to 60 % of the information in the user documents and a comparable classification performance than using all available data. In addition, the results obtained by personal phrases considerably outperformed those from non-personal sentences, indicating their greater suitability for the AP task. We consider these findings could be further applied in the design of strategies for the construction of AP corpora, novel feature selection methods, as well as new feature and instance weighting schemes.

A Novel Criterion to Obtain the Best Feature Subset from Filter Ranking Methods.

The amount of data available in any field is permanently increasing, including high dimensionalities in the datasets that describe them. This high dimensionality makes the treatment of a dataset more complicated since algorithms require complex internal processes. To address the problem of dimensionality reduction, multiple Feature Selection techniques have been developed. However, most of these techniques just offer as result an ordered list of features according to their relevance (ranking), but they do not indicate which one is the optimal feature subset for representing the data. Therefore, it is necessary to design additional strategies for finding this best feature subset. This paper proposes a novel criterion based on sequential search methods to choose feature subsets automatically, without having to exhaustively evaluate rankings derived from filter selectors. The experimental results on 27 real datasets, applying eight selectors and six classifiers for evaluating their results, show that the best feature subset are reached.

Emphasizing personal information for Author Profiling: New approaches for term selection and weighting

Abstract The Author Profiling (AP) task aims to predict specific profile characteristics of authors by analyzing their written documents. Nowadays, its relevance has been highlighted thanks to several applications in computer forensics, security and marketing. Most previous contributions in AP have been devoted to determine a suitable set of features to model the writing profile of authors. However, in social media this task is challenging due to the informal communication. In this regard, we present a novel approach, which considers that terms located in phrases exposing personal information have a special value for discriminating the author’s profile. The aim of this research work is to emphasize the value of such personal phrases by means of two new proposals: a feature selection method and term weighting scheme, both based on a novel measure called Personal Expression Intensity (PEI) which scores the quantity of personal information revealed by a term. For evaluating the latter ideas, we show experimental results in age and gender prediction of media users on six different collections. Average improvements of 7.34% and 5.76% for age and gender classification were obtained when comparing to the best result from state-of-the-art, indicating that personal phrases play a key role for the AP task by means of selecting and weighting terms.

Computer-aided diagnostic approach of dermoscopy images acquiring relevant features

In skin cancer detection, automated analysis of borders, colors, and structures of a lesion relies upon an accurate segmentation process and it is an important first step in any Computer-Aided Diagnosis (CAD) system. However, irregular and disperse lesion borders, low contrast, artifacts in images and variety of colors within the interest region make the problem difficult. In this paper, we propose an efficient approach of automatic classification which considers specific lesion features. First, for the selection of lesion skin we employ the segmentation algorithm W-FCM.1 Then, in the feature extraction stage we consider several aspects: the area of the lesion, which is calculated by correlating axes and we calculate the specific the value of asymmetry in both axes. For color analysis we employ an ensemble of clusterers including K-Means, Fuzzy K-Means and Kohonep maps, all of which estimate the presence of one or more colors defined in ABCD rule and the values for each of the segmented colors. Another aspect to consider is the type of structures that appear in the lesion Those are defined by using the ell-known GLCM method. During the classification stage we compare several methods in order to define if the lesion is benign or malignant. An important contribution of the current approach in segmentation-classification problem resides in the use of information from all color channels together, as well as the measure of each color in the lesion and the axes correlation. The segmentation and classification measures have been performed using sensibility, specificity, accuracy and AUC metric over a set of dermoscopy images from ISDIS data set

Data Mining for Discovering Patterns in Migration

Nowadays, Data Mining has been successfully applied to several fields such as business administration, marketing and sales, diagnostics, manufacturing processes and astronomy. One of the areas where the use of Data Mining has not been well used is in the solution of social problems, where making effective decisions is essential to offering better social programs. In particular, this paper presents an analysis of Migration, which is an important social phenomenon that affects cultural, economic, ideological and demographic aspects of society, among others. This paper is based on an experiment with data processing and clustering analysis of demographic factors related to migration in the State of Hidalgo, Mexico. This study reveals the character and description of clusters obtained with data mining techniques. The knowledge from this characterization is potentially useful to government and social service agencies in the State of Hidalgo for the creation of specific social programs that might be device to mitigate the migration of the population.

Application of Decision Trees for Classifying Astronomical Objects

Data mining techniques used to analyze and discover data and correlations already present in databases, showed to be very reliable and useful especially when large volumes of data are processed. These techniques have been applied to many areas, such as marketing, medicine, diagnosis, business, biology, astronomy and others. In particular, astronomy requires techniques that allow the recognition or classification of astronomical objects, for example galaxies, stars or quasars, from databases that contain millions of objects. Due to this, astronomers often deal with the analysis of large amounts of data obtained from telescopes, seeking for several characteristics for their interpretation. Decision tree is one of the most used techniques in data mining because of its simplicity to explain the results. Besides, there are decision tree algorithms that work with parallel and incremental techniques, which help to process large databases for classifying new objects faster than traditional algorithms. ParDTLT algorithm, which possesses these characteristics, was used in this work in context of astronomical objects catalogue SDSS, with the aim of obtaining decision rules to help astronomers to understand the behavior patterns of different kinds of astronomical objects.

Parallel k-most similar neighbor classifier for mixed data

This paper presents a paralellization of the incremental algorithm inc-k-msn, for mixed data and similarity functions that do not satisfy metric properties. The algorithm presented is suitable for processing large data sets, because it only stores in main memory the k-most similar neighbors processed in step t, traversing only once the training data set. Several experiments with synthetic and real data are presented.