Modeling User's Sentiment in User Segmentations: An Argumentation Approach for User Centered Design

MARÍA PAULA GONZÁLEZ; CARLOS IVÁN CHESÑEVAR; RAMÓN BRENA

LECTURE NOTES IN COMPUTER SCIENCE

Springer

Lugar: Berlin; Año: 2015 vol. 9172 p. 595 - 606

0302-9743

User segmentation is a practice of clustering an audience based on mutually exclusive subsets of individuals that are similar in specific ways. Nowadays user segmentation is crucial not only for the industry but also for the field of User Centered Design, where achieving an accurate understanding of the user?s behavior in the current e-scenario is becoming a complex task. The segmentation could be based on demographic issues, social-economical features, psychographic data, physical characteristics and psychological profiles, etc. This paper proposes a novel strategy for the automatic detection of critical segmentation factors that divide users focused on their feelings and opinions towards a particular topic. Given a topic and on the basis of user?s text-based opinions posted at Web 2.0 services (such as social networks, microblogging platforms, online review systems, online news media, etc.), our proposal introduces an argument-oriented methodology that integrates argumentation theory, sentiment analysis and opinion mining including the computational treatment of incomplete, contradictory or potentially inconsistent information. The mining process is characterized in terms of dialectical analysis of opinions (atomic or more complex opinions constructed by an aggregation mechanism) according to a preference criterion given by topic and feature specificity. As a result, an ?opinion analysis tree? rooted in the first original topic is automatically constructed and visualized, in which any node models a user segmentation, showing the factor that define the segmentation as well as the particularities that group the subset. This way, traditional problems associated with the subjective interpretation of user?s opinions expressed in natural language are minimized. Besides, instead of defining a user?s statistical sample, all available information is considered and possible, not evident critical segmentation factors could be discovered, thus enhancing a rational decision making process.