Last week, we have tried to train a sentiment analyzer and the result is not bad. If you have missed that article, please read it at here: Yiu (2019). However, it seems to be tedious to train the machine learning sentiment analyzer for every new project, also the software is a bit limited for the free-of-charge version.
A real artificial intelligent machine should not be required to be re-trained again and again, because once it has learned, then it should be able to remember forever. Thus, if we can have a pre-trained sentiment analyzer, then we can use it immediately, without doing the training job.
Fortunately, Excel 2013 and afterward provides a free-of-charge pre-trained sentiment analyzer called Azure Machine Learning for an add-in. MrExcel.com provides a youtube podcast to teach it step by step, you may follow at the following link: MrExcel.com (2016).
First, INSERT the add-in AZURE Machine Learning first, then filling in the input data range and the output cell, it will analyze the phrases into either POSITIVE, NEUTRAL or NEGATIVE by a pre-trained lexicon (MPQA Subjectivity Lexicon at http://bit.ly/1SRNevt).
However, the accuracy rate is relatively low. Here (Figure 1) shows the first 20 phrases of my trial, some of them are fatally wrong. It seems to be accurate for POSITIVE sentiment, but very poor in analyzing NEGATIVE ones.
For example, “Rubbish” is considered POSITIVE! but “There is no big mistake” is regarded as NEGATIVE! Worse still, such an encouraging phrase: “don’t give up, work hard next time” is designated as NEGATIVE!
It shows the limitations of a pre-trained sentiment analyzer, and the importance of allowing users to re-train it, otherwise it may do more harm than good. Just like us, we have to have life-long learning.
It is basically the difference between automation and artificial intelligence (Yiu and Yau, 2006). Pre-trained model is an automated model which follows the instructions to carry out the tasks, if the instructions are wrong, then the tasks must not be well performed. AI on the other hand can learn continuously, so that its mistakes can be rectified and the performance can be improved.
However, instead of limiting to just one user to train the machine, if a group of users can share the training tasks, then the machine can learn faster and more. All users can then share the ever-improving pre-trained sentiment analyzer.
(This article is for the Data Intelligence Series)
MrExcel.com (2016) Learn Excel — Sentiment Analysis — Podcast 2062, Youtube. https://www.youtube.com/watch?v=QqA9eQXK3Bk
Yiu, C.Y. and Yau, Y. (2006) ‘A learning model of intelligent home’, Facilities, 24 (9/10), 365–375.
Yiu, C.Y. (2019) Machine Learning Tool for Sentiment Analysis, Medium, Dec 29. https://medium.com/@ecyY/machine-learning-tool-for-sentiment-analysis-f5b354f4264d