Dear Tridib,

It is an interesting problem and dataset. Please find a short list of what is available. In case I come up with any more concrete steps, I will update my answer.

Wikipedia:

http://en.wikipedia.org/wiki/Multi-label_classification

A study:

"Multilabel Text Classification for Automated Tag Suggestion" Ioannis Katakis, Grigorios Tsoumakas, and Ioannis Vlahavas

Software:

http://scikit-learn.org/stable/modules/multiclass.html

http://scikit-learn.org/stable/auto_examples/plot_multilabel.html

http://www.nltk.org/api/nltk.classify.html#nltk.classify.api.ClassifierI.classify

Extracting features from text: 

http://scikit-learn.org/stable/modules/feature_extraction.html

Greetings,

Ali

Great...Thanks a lot Ali. I will go through these suggestion carefully.

@Ali, Can you access the http://scikit-learn.org/stable/auto_examples/plot_multilabel.html  link you provided. Every time, I am trying to access, sourfeforge gives an error: saying the project is blocked due to bandwidth over-usage...Is there any other way to see the content..

Thank you...

Dear Tridib,

I can access the link now. There was a problem on the web site in these days, but it seems to be solved.

Greetings,

Ali

This tool can also be useful:

https://gate.ac.uk

Since Ronaldo already mentioned the GATE framework, which is implemented in Java as opposed to the Python tools mentioned by Ali... In GATE you actually use other tools to accomplish machine learning tasks. One of those is WEKA

http://www.cs.waikato.ac.nz/~ml/weka/ .

WEKA can also be used for text classification on its own

http://weka.wikispaces.com/Text+categorization+with+WEKA .

It includes preprocessing algorithms like TF-IDF as well as classification algorithms like SVN or naive bayes.

I'd also like to recommend two books explaining text mining and machine learning since you said you are new to the topic:

[1] R. Feldman and J. Sanger, The Text Mining Handbook. Cambridge University Press, 2006.

[2] J. Han, M. Kamber, and J. Pei. Data Mining: Concepts and Techniques. Morgan Kaufmann, 3rd edition, 2011.

Hope this helps and good luck classifying!

Thorsten

Thanks to both Thorsten and Ronaldo for useful information. I have used WEKA before, but not for text classification. However, I am required to use python/C as the main language. I am using python. I will check out the GATE framework. 

Thanks for the information regarding the books. Thanks a lot once again. 

@Ali, thanks a lot. I could access the content. Thanks a lot for taking the time to answer my queries.

However, I have another query if you could contribute to the answer that would be great. 

Currently, the list of company names are fixed. However, in reality, this isn't the case. Rather, it is usually the case that new companies form and some goes out of business. Thus the company names list is dynamic. If we are faced with such a situation, how do you go about solving it. Any thoughts ?

Best,

Tridib

Dear Tridib,

You can use a machine learning approach. Learning the left and right context, and properties of company names will enable the system recognise new companies. This can be specific for new formed and bankrupted companies as well. For example new companies will be announced in a specific way, it is the same for bankrupted companies. A study:

"Named Entity Recognition using an HMM-based Chunk Tagger" GuoDong Zhou Jian Su, Link: http://dl.acm.org/citation.cfm?id=1073163

A presentation: http://www.dfki.de/~neumann/esslli04/reader/ie-lec3-1.pdf

You can use Wikipedia since it will be updated at least for the new company names and will contain the relevant information about bankrupted companies: http://www.dfki.de/~neumann/WikiMiningSeminar2009/presentations/named-entity-presentation.pdf

You can define information extraction rules to identify new formation and bankrupt information as well: https://github.com/machinalis/iepy

An automatic method:

"Generating Pseudo-ground Truth for Predicting New Concepts in Social Streams" by David Graus, Manos Tsagkias, Lars Buitinck, Maarten de Rijke

A note: Company names will contain ambiguous terms such as "apple", "puma". Therefore, you may need to use a more sophisticated method than a list look up.

Let me know in case you think I can contribute more ;)

Good luck!

Ali

@Ali, thanks many times for your updates. These are my first forays into NLP and I learnt a lot from your updates.

Yes, I believe the I need a sophisticated algorithm here. Lets see how that pans out.

Thanks once again.

Best,

Tridib

Kindly let me know what is the language of of your text first? Because some built in tools are available in many language domains.

Try GATE, UIMA, TextMarker, OpenNLP, Mathematica, etc. Each one has its strength and weakness. 

@Qaiser, thanks. The language is English. I need to use python. 

@Justin. Thanks for your suggestions.

Have you read this book: http://www.nltk.org/book/

You should probably test approaches combining dictionaries (gazetteer) and lexico-syntactic or lexico-semantic patterns (like JAPE finite state transducer).

I dont know if 2000 exemples are enough, but you can try using patterns learning / recognition approaches.

@david thanks a lot. Looking at the book you mentioned.

I would first consider whether AI is needed at all - maybe you could use the regular expression features in Perl.

If you do need AI take a look at semantic similarity measures for words, terms or short text. You may be able to use these for relevance level by measuring the similarity between an exemplar piece of text and text in your articles. See

Sentence Similarity Based on Semantic Nets and Corpus Statistics

August 2006 (vol. 18 no. 8)

pp. 1138-1150

Yuhua Li, David McLean, Zuhair A. Bandar, James D. O'Shea, Keeley Crockett, "Sentence Similarity Based on Semantic Nets and Corpus Statistics," IEEE Transactions on Knowledge and Data Engineering, vol. 18, no. 8, pp. 1138-1150, August, 2006.

 Yuhua Li

David McLean

Zuhair A. Bandar

James D. O'Shea

Keeley Crockett

You should be able to get the last author's pre-publication version on my researchgate pages.