Twitter sentiment analysis using a Deep Learning appraoch

[lawso...@gmail.com]

Hello There 

i am Student and i did my finale year project on sntiment analysis on Twitter using deep learning 

so i used the Sentiment140 dataset 

i get a good reuslts with a high accuracy 

that equals to 0.93 for an LSTM model

and equals to 0.92 for and CNN model

so any one who needs help or my small experience 

i am at his service 

[vera...@gmail.com]

Hi lawso

I am studying the dataset of Sentiment140 and the dataset seems to be missing the 2 (neutral) data, and is not encoded utf-8.

http://help.sentiment140.com/for-students/

Any possibility to update it? 

Thanks.

[M KA]

hello 

veralillo

yes in their paper the mentioned that the dataset contains three classes 0,2 and 4 but  actually it contains only the 0 and 4 classes 

so i did a binary classification 

if you have a problem in the dataset or you want another link to it 

tel me and i will provide the link to get it 

 

--
You received this message because you are subscribed to a topic in the Google Groups "Sentiment140" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/sentiment140/qBuJ32xZ7oc/unsubscribe.
To unsubscribe from this group and all its topics, send an email to sentiment140...@googlegroups.com.
To post to this group, send email to sentim...@googlegroups.com.
Visit this group at https://groups.google.com/group/sentiment140.
To view this discussion on the web visit https://groups.google.com/d/msgid/sentiment140/e13d30ac-5341-4a9d-949b-4537df0605d5%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[tobiasnu...@gmail.com]

Quite impressive regarding the fact that state-of-the-art approaches on this dataset achieve around 80% accuracy [1].

Please enlighten us and share your knowledge ...

[chaule]

Hi,

That's stunning! It would be even better if you can show us how to reproduce the results (data preprocessing, model architecture, hyperparams, train-test split ratio, evaluation metrics...) and share your model.

Thanks all the same!

[obsi...@gmail.com]

how you prepare dataset for local languagee/native languages?if you have dataset send me sample

[lizhenz...@gmail.com]

Hi, are your reported results tested on the given clean test set of sentiment140? 

What is your neural network architecture?

Many thanks!

[M KA]

Hello there
as we know in the learning process we have training set and test set

so i divide the dataset into 67% for the training and the rest for the test

for the neural network architecture i build two models  ' LSTM model +  CNN model '

--

You received this message because you are subscribed to a topic in the Google Groups "Sentiment140" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/sentiment140/qBuJ32xZ7oc/unsubscribe.
To unsubscribe from this group and all its topics, send an email to sentiment140...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/sentiment140/b0c6ace3-7998-46a9-87a2-128818662a1d%40googlegroups.com.

[vera lee]

Thanks for your prompt response.

Several more questions:

Why don't you use the given clean test dataset to evaluate the model?

Do the neural models perform better than the traditional machine learning models published in the paper " Twitter Sentiment Classification using Distant Supervision"?

Thanks for your time!

Happy new year!

Vera

To view this discussion on the web visit https://groups.google.com/d/msgid/sentiment140/CAAtW_iA1y4Y2uivRFO1EYawi1mbHTS81YUU05BXzUBKKxk2yuA%40mail.gmail.com.

[M KA]

Hello again

yes i understand your point but in my project i applied the cross validation methods (Leave one out k fold cross validation.....) that is why i used the whol dataset with 1.4 M tweets

and yes it did give a much more great results then traditional machine learning models mentioned in the paper i get an accuracy up to

0.93 for an LSTM model and equals to 0.92 for and CNN model

To view this discussion on the web visit https://groups.google.com/d/msgid/sentiment140/CAKgAbXuM1SxABpXTDBNZMxcUS2-jAq92DcAdR1NbmvU0oBH3fg%40mail.gmail.com.

[vera lee]

Do you use some pre-trained word embeddings or use some important parameter settings?

When I trained the simple CNN model with only one Conv1D layer and used 1% of the training data as validation data, the validation accuracy is at most 82%, which is much lower than your CNN model.  What is the architecture of your CNN model?

To view this discussion on the web visit https://groups.google.com/d/msgid/sentiment140/CAAtW_iB%3DELnzEWtJJtM%3DB7YWE_nhNCfjFKf6tJ9pi_g%3Da34Nkw%40mail.gmail.com.

[M KA]

Hello

yes i did use word Embedding using Glove here it is a link

https://nlp.stanford.edu/projects/glove/

for the  architecture :

i used Embedding layer

Conv1D layer that have a MaxPooling1D layer following it

also you have to use a dropout layer between the different layers

in the end add a dense layer

for the optimizer i used adam optimizer

and the binary_crossentropy for the loss function

To view this discussion on the web visit https://groups.google.com/d/msgid/sentiment140/CAKgAbXv8c8Wq5ORnVurMJt6cQkks4ERP6EP%3DgwiBcHXGVoSktQ%40mail.gmail.com.

[sima.s...@gmail.com]

hello 

i am student too and i should show an execution of this project to my teacher .i were happy if you help me how could i run this dataset with knime software.

[M KA]

Hello there 

The dataset is on csv format

I find this great tuto on how to read csv files using this software  me

https://youtu.be/Y5EnUjZ-R-s

--

You received this message because you are subscribed to a topic in the Google Groups "Sentiment140" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/sentiment140/qBuJ32xZ7oc/unsubscribe.
To unsubscribe from this group and all its topics, send an email to sentiment140...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/sentiment140/fe61b8e9-f0a7-4274-9980-2f028ceabc39%40googlegroups.com.

[Rony Sharma]

Dear Lawso,

How are you? Hopefully you are doing well. You are complete Final year Project  on Sentiment Analysis on Twitter using Deep Learning. Your Project Accuracy is to good. Also you Use LSTM Model and CNN Model. I am also Student of Final Year Student. If you Don't Mind, Give me your Paper and so on. Also LSTM And CNN Model Code. Thank you for Advance.

Rony Sharma 

[ikramh...@gmail.com]

Le lundi 3 juin 2019 15:37:42 UTC, lawso...@gmail.com a écrit :

Hello There 

i am Student and i did my finale year project and i need thos data set to complete can you help me because I didn't  find it 

[M KA]

Hello there 

Yes, you can get the data from this link 

https://drive.google.com/file/d/1BT9c_w3BxQ1JwZOfXunStGU9ylN3NlUk/view?usp=drivesdk

--

You received this message because you are subscribed to a topic in the Google Groups "Sentiment140" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/sentiment140/qBuJ32xZ7oc/unsubscribe.
To unsubscribe from this group and all its topics, send an email to sentiment140...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/sentiment140/087813b9-1415-4210-8129-64674632629c%40googlegroups.com.