hi everyone ,
I'm working on a textmining research and need to know how can i prepare a labeled dataset for supervised machine learning to extract and detect aspects?
for example my sentence is : "this phone is amazing, long battrylife and good price, the only bad thing is it's color."
as you know "phone", "batterylife", "price" and "color" are aspects.
how can i label these words ?
as an example use Weka and SMO function for predecting
The labeling approach is based on your research. You can not label your data blindly. Two labeling methods can be used, though:
- labeling the data based on a certain background.
- labeling the data based on a predefined dictionary (i.e., language-based dictionary) where weight can be assigned to each word according to a particular dictionary. Then, summing all the weights to get the final weight of the sentence--the label of the sentence. Using this approach, a label of 3 is going to be assigned to your sentence (this phone is amazing, long battrylife and good price, the only bad thing is it's color) and thus its label is considered to be a 'positive' label.
Once you have your data labeled, WEKA can be used via SMO to learn your labeling and make future predictions.
HTH.
Dr. Samer Sarsam
I agree with the answer of Dr. Samer, as it is a straightforward academic answer.
Fernando
thanks All.
but as Dr. Samer said , "posetive" tag for this sentence just set polarity in sentence level.
I wanna Detect Aspects.
another example to explain my question :
look at these sentence below and the tag for each one
1. "I buy it last week and it's satisfying me", OVERALL
2. "good memory hard chip" , TECHNICAL
4. "a little high price ", COST
5. " best cpu I've ever seen", TECHNICAL
6. "so expensive phone", COST
7. "I like it", OVERALL
.................
I labeled all my sentence as this order and my problem is i can't tag a complex sentence which has more than one aspect
as an example : "very good price and display"
You're welcome, Mostafa.
Maybe if you have more than one aspect, ' OVERALL ' is reasonable. Another option is to add a new category to deal with the multiple aspects issue.
Dr. Samer Sarsam
- Badges
- Science topic