Foundations of uncertainty management for text-based sentiment prediction

Abstract

Analyzing the sentiment of Social Networks users is an attractive task, well-covered by the Sentiment Analysis research communities. Alongside, predicting the rating/opinion of users in Social Networks or e-commerce platforms is another attractive task covered by the Recommender Systems research communities. However, there is a rather new field of study that takes advantage of both of the mentioned scopes to predict the “unexpressed” opinion of users, based on their written sentiments and their similarity. Although the Social Network extracted data (due to the sparsity of the addressed items by different users) deals with high volumes of uncertainty, none of the few dozens of conducted studies in the Sentiment Prediction field focuses on managing the mentioned uncertainty. In this dissertation, we introduce the necessary foundations for constructing an Uncertainty-handling Sentiment Prediction system, by means of possibility theory, fuzzy theory, and probability theory. Moreover, we define an international project called probabilistic/possibilistic Text-based Emotion Rating (pTER) to fill and then enrich the gap of uncertainty management in Sentiment Prediction. pTER comprises two sub-projects: Scalar and Interval pTER. This dissertation provides five foundational research studies in the scalar pTER. Although the mentioned studies are sufficient for the targeted system, we let the scalar pTER system, itself, to be disseminated only after it can use its entire potency by utilizing the in-progress research projects of the other researchers of the pTER project, defined by this dissertation. In addition to the presented scalar-pTER studies, we also propose one research study in the interval pTER project which goes one step further in Uncertainty-handling and takes the measurement errors of the scalar pTER sub-systems into account. The presented studies in scalar- and interval-pTER belong to three phases: (I) Uncertainty-handling NLP platform, (II) Uncertainty-handling Sentiment Analysis, and (III) Uncertainty-handling Collaborative Filtering. The conducted experiments in this dissertation prove the superiority of our Uncertainty-handling approaches in all of these phases, in comparison to the corresponding state-of-the-art

Hay un campo de estudio bastante nuevo que aprovecha análisis de sentimiento y filtrado colaborativo para predecir la opinión "no expresada" de los usuarios, en función de sus sentimientos escritos y su similitud. Aunque de la red social se extraen datos (debido a la escasez de los elementos tratados por diferentes usuarios), abarcando e un alto volumen de incertidumbre, ninguna de las pocas docenas de estudios realizados en el campo predicción del sentimiento se centra en la gestión de la incertidumbre mencionada. Presentamos los fundamentos necesarios para construir un sistema de predicción del sentimiento de manejo de la incertidumbre, mediante la teoría de la posibilidad, la teoría difusa y la teoría de la probabilidad. Además, definimos un proyecto internacional llamado Probabilistic / Posibilista basado en el Emoción Rating (pTER) de textos para llenar y luego enriquecer el área de investigación de la gestión de la incertidumbre en predicción del sentimiento