This is a repository for a university project. The task is to create a machine learning model to predict the sentiment of a tweet about airlines based on annotated data. We use the readme for brainstorming.
ideas:
FE extraction from RAW
- UPPERCASE
- emojis
- punctuation signs (!;?;...;..)
- text length
- Timestamp
CountVectorizer bigrams y trigrams
- Ignoring case
- Ignoring punctuation
- stopwrods
- Lematizción
- We can use the labels to obtain more suitable vocabulary, maybe neutral comments are misleading.
- We need to do crossvalidation at some point
- The order of words can change sentiment
- For next class (17/04/2018) we have to upload to kaggle results with a most suited model
- We will need to use feature selection [we will see it in next classes]
- Naive Bayes only considers if the word is in the text, it does not consider number of appearences
- At the end of the project (03/07/2018), we will need to present a news repport talking about some airline companies
-
15 de June is the deadline to present boths kaggles!
-
Tenemos que hacer una web/blog con una explicación técnica de lo que hemos hecho --> 3 Julio. Memoria. Que sea didactico. Link al github con el codigo si es posible.
-
Datacleaning
-
exploración de datos
-
machine learning...
-
El metodo que hemos hecho para presentar el kaggle
-
Además de esto ver qué historias podemos extraer sobre las aerolineas.
-
-
3 Julio haremos la presentación con un ppt o similar. 2-3 min en el programa que hemos hecho + 7 min con las preguntas/conclusiones y preguntas que nos hemos hecho.
Models
-
Naive bayes
-
SVM-RBF
-
RF
-
Boosting
Visualization
- distribution
How to handle the unevenly distributed dataset? See percents:
- negative 0.629668
- neutral 0.210155
- positive 0.160178
