Hi. I have a query regarding Text Classification. I have a list of words with the following attributes. word, weight, class. The class can be positive or negative. Weight is between -1 to 1. How can I train a classifier like SVM using this word list to classify unseen documents? An example in any tool is welcome
WEKA is a suitable tool for your task. You can use WEKA's document processing and classifier modules.
Hmm, maybe I am misinterpreting your data, but how can a positive class have a weight = -1 and vice versa?
Anyways I suggest Rapidminer, WEKA, KNime and Orange. R and Python if you want to program yourself.
Why do you want to train a SVM classifier? I suggest you should begin with a simple classifier like Naive Bayes to see how it works. Like Kasper Christensen, I don't really understand your data either. Each word has a weight or each document has a weight?
It seems you are using sentiment lexicon to classify the document. First you need to extract the attributes based on your lexicon. There are some ways, but the easiest is by extracting two attributes, call it as posScore and negScore that represent score sum of positive words and negative words consecutively, for each your unseen document. Then do classification.
The other way, you also can consider, the minimum, maximum, average score for both type.
Why don't you have a look at the RapidMiner tool. It has a very nice visual interface, handling is quite easy to learn and there are several examples and videos (on youtube etc.) showing typical analysis cases for SVMs.
See https://rapidminer.com/ for more details.
Kind regards
Christian
For the supervised text classification, you have to start by creating the dictionary also called vocabulary. In the case of the bag of word representation, the dictionary contains the words (without repetition) of the entire training corpus. Then each document is a vector where the size of it is equal to the size of the dictionary (each element of the vector represent a word of a document). Then you feed the set of training vectors to one of the machine learning algorithms that you want to use. If you want to use SVM, you can use the libsvm package. After this, a learning model will be created. This one will allow you to predict the class of the new documents.
Best regards,
Riadh Belkebir
Thankyou all for your replies. @Riadh Belkebir this is exactly what I am looking for. LIBSVM-Light would do the job. I just have to provide the training/test dataset in its required format.
http://svmlight.joachims.org
weka , rapid miner, sklearn python library are the easy to use for classification
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts