[Anubad] Project to Develop a system with multi-lingual capabilities in order to receive answer to user specific queries

[Abhishek Gupta]

Sorry for the long mail. But any comments anyone so that I can work out
further.

Regards
Abhishek

On 19 March 2012 11:47, Abhishek Gupta <abhishekgupta.iitd at gmail.com> wrote:

> Hi Sankarshan,
>
> Thanks for the reply. In my view, possible input sets are as follows :
>
> *1 - Model set of questions and answers.*
> In this case, if a user asks a question we can find the co-relation of the
> question with any of the existing solution(s). On the basis of the
> questions shortlisted, now we have some idea what answers to use for the
> same. Next step, will be to answer the asked question based on answers of
> the model questions shortlisted.
>
> The advantage of the approach is that most of the people already have a
> model set of questions and answers with them, because of which the approach
> won't need much input from the administrators.
>
> *2 - Ontology database*
> The other more obvious, but less convenient option is to construct an
> ontology or a semantic network of the information. Next we try to shallow
> parse the users' question, followed by semantic analysis (like mapping to a
> prolog query on our ontology database, finding the answer and then using
> the same to reply the user in a more 'humane' way. This is quite similar to
> the model that SIRI is used.
>
> *3 - Unstructured data like wiki page*
> This approach even way more convenient to the user, might not generate
> really good results. Over here we give unstructured data like say a wiki
> page, the system proposed should search the page to find the answer, and
> accordingly modify the same so that it looks socialable and relevant to the
> question asked.
>
> Note:
> We can integrate the solution with the existing bot code-base like *alice*,
> so as to take advantage of the extensive knowledge base created.
>
> Like in the bot https://www1.paypal-virtualchat.com/, we can try to
> orient the discussion by asking questions from the users with *constrained
> inputs like extracting a boolean value* from the answer to see if we are
> trying to answer indeed helps us.
>
> The project can be done in two *phases*. In first phase we can focus on
> more *objective *questions, while on other phase *subjective *questions
> can be stressed upon.
>
> Please add if I am missing something or wrong at some place. There are
> some really good publications with respect to the same specially in context
> of SIRI. Finally, I wanted to ask about the possibility of a research *
> publication* if we are able to achieve some noteworthy results, which I
> am quite confident of achieving.
>
> Thanks and Regards
> Abhishek Gupta
> 3rd Year, Computer Science, IIT Delhi
> abhishek.cc
>
> On 19 March 2012 10:53, Sankarshan Mukhopadhyay <
> sankarshan.mukhopadhyay at gmail.com> wrote:
>
>> Hi Abhishek,
>>
>> I have copied the mailing list in the reply and, have responded to the
>> relevant bits of your email.
>>
>> On Sun, Mar 18, 2012 at 4:25 AM, Abhishek Gupta
>> <abhishekgupta.iitd at gmail.com> wrote:
>>
>> > I am a student of 3rd year pursuing a B.Tech degree in Computer Science
>> at
>> > IIT Delhi (Indian Institute of Technology, New Delhi, India). I read
>> about
>> > the project ideas for Ankur and would be really glad if I can work on
>> the
>> > project "Develop a system with multi-lingual capabilities in order to
>> > receive answer to user specific queries" under your guidance to
>> complement
>> > my present knowledge.
>>
>> It is good to see someone have an interest in this specific project.
>> To answer the question you had asked in a separate email, I'd like the
>> proposal from you to consider the training data for this system along
>> with the rationale on why you selected that as a training data. The
>> way we would like to undertake the projects is provide a general sense
>> of direction and thereafter request the interested candidates to nail
>> down the objectives and the route as well as the metrics which would
>> help us gauge the success/failure of the projects with various
>> milestones.
>>
>> /sankarshan
>>
>> --
>> sankarshan mukhopadhyay
>> <http://sankarshan.randomink.org/blog/>
>>
>
>