INSTITUTO POLITÉCNICO NACIONAL CENTRO DE INVESTIGACIÓN EN COMPUTACIÓN Laboratorio de Procesamiento de Lenguaje Natural Polarity summarization with opinion mining

This thesis work presents a novel framework for Opinion Summarization. This task consist in the extraction of parts of a opinionated text that represents the main opinions and build a legible and human-readable opinion summary. This task is very difficult and it involves several methods related to Opinion Mining, Natural Language Processing, Machine Learning, Information Retrieval and other areas. The main goal of this work is to expand the state of the art in Opinion Mining by researching in new ways for Opinion Summarization and the tasks it involves. Another goal of this work is to provide to the scientific community new tools to tackle challenges in this area. At the moment of writing this thesis there are few, if any, public available opinion summarization tools. The present work explains in detail the models that describes the phenomena found investigating this task, the different methods used and/or developed for Opinion Summarization, their implementation and the results obtained. An opinion summarization tool receives an opinionated text document. The output could be the sentences or phrases of the given document that summarize better the opinions found in the text. Another possibility would be some sort of structured summary describing the quantitative properties of the opinions found in the document. This work uses both approaches. The proposed method for opinion summarization can be described in 3 steps: • First it is necessary to decompose the input document into sentences and determine which of these sentences are opinionated, obtaining their polarity in the process.