Publication: Análisis de sentimientos en Twitter en castellano con redes neuronales recurrentes LSTM
Loading...
Identifiers
Publication date
2021
Defense date
2021-05
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
Desde hace varios años, el crecimiento y el impacto de las redes sociales en la sociedad ha crecido en gran medida. Hoy en día, tanto a las grandes empresas como a los particulares, les interesa estar en constante contacto con las redes para saber lo que está pasando en el día a día. En el caso de las empresas, este interés viene motivado, entre otras cosas, por la posibilidad de obtener cierta información que pueda ser beneficiosa para las mismas.
Los usuarios de las redes sociales están constantemente redactando opiniones, comentando películas, libros y un sinfín de opciones. Todo esto puede ser utilizado por las empresas para rescatar información y usarla de manera que pueda ser rentable de cara a campañas, anuncios, etc. Con la finalidad de estar siempre por encima de la competencia.
Una de las técnicas que permite a dichas empresas obtener esa información es el análisis de sentimientos en las redes sociales. Este TFG pretende abordar este problema de procesamiento de lenguaje natural mediante el desarrollo de un algoritmo capaz de realizar análisis de sentimientos en la red social twitter. Más concretamente en los tweets que involucren a la empresa de telefonía Yoigo (MásMóvil).
El algoritmo será capaz de realizar una polarización de los sentimientos (positivo, negativo y neutral) en cada uno de los tweets que se introduzcan. Esto permite, entre otras cosas, hacer estudios para saber las épocas en las que los tweets son más positivos, cuándo hay un repunte de tweets negativos, el porqué de ello, etc.
El código junto con los resultados se puede encontrar en el siguiente link:
https://www.kaggle.com/pedrohernandezelp/twitter-sentiment-analysis-lstm-with-traductor
Durante el estudio de este proyecto se han encontrado numerosas maneras de resolver el problema, sin embargo, no todas revierten los mismos resultados. El algoritmo elegido para resolver este sistema son las Redes de Neuronas Recurrentes LSTM y más adelante se estudiará el por qué esta es la mejor manera de resolverlo.
For several years now, the growth and impact of social networks on society has been increasing. Nowadays, both large companies and individuals are interested in being in constant contact with the networks to know what is happening on a daily basis. In the case of companies, this interest is motivated, among other things, by the possibility of obtaining certain information that can be beneficial for them. Users of social networks are daily writing opinions, commenting on films, books and a host of other options. All of this can be used by companies to gather information from users and use it in a way that can be profitable for campaigns, advertisements, etc. With the aim of always being ahead of the competition. One of the techniques that allows these companies to obtain this information is sentiment analysis in social networks. This TFG aims to address this problem of natural language processing by developing an algorithm capable of performing sentiment analysis in the social network twitter. More specifically in tweets involving the telephone company Yoigo (MásMóvil). The algorithm will be able to perform a polarisation of the sentiments (positive, negative and neutral) in each of the tweets that are entered. This allows, among other things, studies to be carried out to find out the times when tweets are most positive, when there is a spike in negative tweets, the reason for this, etc. The code and results can be found at the following link: https://www.kaggle.com/pedrohernandezelp/twitter-sentiment-analysis-lstm-with-traductor By studying the problem, numerous ways have been found to solve the problem. However, not all of them have the same results. The algorithm chosen to solve this system is the LSTM Recurrent Neuron Networks and later on we will study why this is the best way to solve it.
For several years now, the growth and impact of social networks on society has been increasing. Nowadays, both large companies and individuals are interested in being in constant contact with the networks to know what is happening on a daily basis. In the case of companies, this interest is motivated, among other things, by the possibility of obtaining certain information that can be beneficial for them. Users of social networks are daily writing opinions, commenting on films, books and a host of other options. All of this can be used by companies to gather information from users and use it in a way that can be profitable for campaigns, advertisements, etc. With the aim of always being ahead of the competition. One of the techniques that allows these companies to obtain this information is sentiment analysis in social networks. This TFG aims to address this problem of natural language processing by developing an algorithm capable of performing sentiment analysis in the social network twitter. More specifically in tweets involving the telephone company Yoigo (MásMóvil). The algorithm will be able to perform a polarisation of the sentiments (positive, negative and neutral) in each of the tweets that are entered. This allows, among other things, studies to be carried out to find out the times when tweets are most positive, when there is a spike in negative tweets, the reason for this, etc. The code and results can be found at the following link: https://www.kaggle.com/pedrohernandezelp/twitter-sentiment-analysis-lstm-with-traductor By studying the problem, numerous ways have been found to solve the problem. However, not all of them have the same results. The algorithm chosen to solve this system is the LSTM Recurrent Neuron Networks and later on we will study why this is the best way to solve it.
Description
Keywords
Procesamiento del lenguaje natural, Redes neuronales recurrentes