Desarrollo de un sistema de detección de necesidades de atención municipal en base al análisis de tweets.

Abstract

Tweets gained fame in recent years due to the use of social networks, they allow analyze different aspects such as feelings and extract topics from them. This project details the development of a needs detection system regarding the Municipality of Quito based on the analysis of tweets. The topics that are involved in this project are the extraction of topics to discover the most important aspects which are spoken in Quito and according to their sentimental polarity classify them as needs and the analysis of feelings to classify the polarity (positive, negative or neutral) of citizenship attitude in each topic found along with the location of important and most mentioned places by citizens. In the extraction of topics, cleaning and data processing techniques were considered, such as eliminating punctuation marks, converting to lowercase, removing stopwords from the Spanish language, among others. Likewise, the algorithm used was Gibs Sampling algorithm for the Dirichlet Multinomial Mixture model (GSDMM), which performed better than the Latent Dirichlet Allocation (LDA) algorithm. After applying GSDMM, seven topics were obtained, of which the sentimental polarity was analyzed. In the sentiment analysis section, a Logistic Regression model was trained to obtain the polarity (positive, negative and neutral) of each topic. In addition, subtopics of some topics were generated due to their low granularity of detail.