Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/124106
Title: Análisis de sentimiento en tiempo real de mensajes de Twitch con algoritmos de clasificación
Author: Sáenz Rubia, Raúl
Tutor: Diego Andilla, Ferran
Others: Ventura, Carles  
Abstract: Sentiment analysis is a subfield of Natural Language Processing whose objective is to analyze subjective texts and extract useful information about the author's polarity of feeling at the document level, of a specific phrase or about an aspect of an product or service of interest for the analyst. Specifically, social networks and micro-blogging pages such as Twitter or Twitch have become inexhaustible sources of texts that can be used to instaltly analyze public opinion on aspects as varied as politics, economy or any other product or service. Twitch, in particular, is a streaming platform that has grown tremendously in recent years, which makes it necessary to improve interaction with the public to get rid of the competition, and Twitch chat is a source of information that can help creators of content to know the opinion of your audience. However, there are currently no extensions for sentiment analysis in Spanish for this platform, so we see an opportunity to innovate.In this work, we use MLP and LSTM deep learning methods to develop a multi-class sentiment analysis classifier with three levels of polarity: positive, neutral and negative, with achieved average results of 75% in accuracy, which is a quite acceptable outcome considering the obstacles found in the analysis such as language ambiguity, unawareness of context and intentional or unintentional misspellings.
Keywords: LSTM
MLP
machine learning
sentimental analysis
natural language processing
neural networks
Document type: info:eu-repo/semantics/bachelorThesis
Issue Date: 20-Jan-2020
Publication license: http://creativecommons.org/licenses/by-nc-nd/3.0/es/  
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
rsaenzTFG0120memoria.pdfMemoria del TFG6,9 MBAdobe PDFThumbnail
View/Open