A user is describing a task to be performed by some company:

I want to move 10 boxes (30x40x50cm) of books and clothes from London to Bristol next week. How much would that cost?

Based on this description, a library/algorithm classifies this as a task probably for removal/transportation company operating preferably in UK. Based on that guess an application queries some search engine for companies belonging the guessed category (e.g., Google maps).

Are there libraries/algorithms that makes such kind of text categorization? I assume the taxonomy of service categories (transportation, translation, etc.) is given by me.

What would be prerequisites? I would prefer already trained algorithm (as I do not have a corpus of categorized task descriptions) or unsupervised one (that does not require such a corpus).

Evaluated candidate solutions

• Identify to which WordNet Domain words in description belong and find prevalent one. Map manually WordNetDomains to company categories and use this mapping to find category of the task.

• uClassify, but my task has been classified as "Home (62.7 %)", "Games" (15.1 %), and "Arts" (13.2 %).

• Textimate.me classified it as "Science & Environment"

• Textwise returned "Business/Consumer_Goods" and "Services/Clothing Arts/Design/Fashion",

The results are rather discouraging. I think the services I tried have the taxonomy close to what I would use, but the text I submit for classification is very short and there is no way for the algorithm to differentiate what is more important for the user: boxes, clothes or maybe moving them? I think a good heuristics for service category could be to focus on verbs rather than nouns, and on places (London/Bristol) in case of area identification.

Still, I believe the only situation here would be to collect a corpus of such tasks and their categories from some online place (where people looks for suppliers to perform such task for them) and machine learn on that. Anyone tried?