Machine Learning technique for learning string patterns - machine-learning

Question

I am new to machine learning and I am looking for a technique to learn string patterns based on a training data set.
My problem:
I have different types of words, belonging to different categories. Each category has somekind of its own pattern (for example one has a fixed length with only special characters, another exists of other characters which only occur in this category of "word").
For example:
"ABC" -> type1
"ACC" -> type1
"a8 219" -> type2
"c 827" -> type2
"ASDF 123" -> type2
...
I am searching for a machine learning technique to learn these pattern on its own, based on training data. I already tried to define some predictor variables (for example wordlength, number of special characters, ...) on my own and then used a Neural-Networks to learn and predict the category. But thats acutally not what i want. I want a technique to learn the pattern for each category on its own - even to learn patterns which I never thought about.
I want give the algorithm the learning data (consisting of the word-category examples) and want it to learn patterns for each category to predict the category from similar or equal words later in production.
Is there a state-of-the-art way to do it?
Thanks for your help

Answer 1

Since you have tag weka, the process would be
1. Create the feed the arff file
Example
#relation weka_mymodel_model
#attribute text string
#attribute ##class## {type1,type2}
#data
'boy am I stupid. I mean, wow, that was a major oversight. let\'s blame it on monday.',type1
..... all your data
2. Load the file in weka software
In the pre-processing tab you can filter (transform) the data; to for example a StringToWordVector that can be used with J48 classifiers etc, but we will leave this for now and only use classifiers that can handle directly your input
3. Classify
In tab "Classify", select attribute ##class## and then select a classifier that can support text directly a good start is the NaiveBayesMultinominal
In the interface of the classifier, set you settings, Stemmer, StopWords, Tokenizer etc.
The classifier to use and with which settings depends on the data, but you can test run the classifier either on "Using training set", "Supplied test set" or "Cross-fold" to understand what outcome your different settings have.
4 Create the model
When you are happy with your settings, export the model (right click the result>>Save model).
5 Use the model
Load the model in java, create the Instance, pass it to the model and retrive your result.
Conclusion
The weka software lets you test different classifier algorithms with different settings, the best way to find the best classifier is to test run the different classifier (use filters, select attributes etc) with different settings on a "Supplied test set" and check outcome.
]