I have adapted a sentiment analysis of text model from the book Codeless deep learning with KNIME by Melcher and Silipo to analyse a different data set.
It reurns the same result for all text examples even those which should have a different result. I have checked through and cannot find the cause of the issue.
Can anyone help please?
Results should show 0 or 1 for sentiment.
https://hub.knime.com/stockdaa/spaces/Public/latest/Sentiment_Analysis_Training_With_Integrated_Deployment%201%20ExcelV01(5)%202
https://hub.knime.com/stockdaa/spaces/Public/latest/Sentiment_Analysis_Deployment_with_Call_Workflow%201%20ExcelV01(5)%202
Hello @stockdaa,
I was unable to execute your workflow because the data file is stored locally on your system. To make your workflow more portable and easier for others to use, consider using relative paths and storing the data alongside the workflow.
I will check the workflow and get back to you.
Best,
Keerthan
2 Likes
Hi @stockdaa,
Training a deep learning model with only 56 records is generally not advisable due to several important considerations like overfitting of model and lack of generalization.
I recommend using simpler classification models like Random Forest. You will find examples of them on the Community Hub.
Best,
Keerthan
2 Likes
Workflow and data below:
knime://knime.mountpoint/Users/stockdaa/Public/Sentiment_Analysis_Deployment_with_Call_Workflow%201%20ExcelV01(5)%203
knime://knime.mountpoint/Users/stockdaa/Public/Sentiment_Analysis_Training_With_Integrated_Deployment%201%20ExcelV01(5)%203
Carillion DB.xlsx (86.8 KB)
Hi @stockdaa,
As mentioned above, when working with a small dataset, it is efficient and effective to use simpler models like Random Forests or employ transfer learning with pre-trained neural networks.
These will help to avoid overfitting and can often deliver better performance.
Best,
Keerthan
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.