I am planning to develop a Q&A system to chat within a limited context of topics. My data is limited so I will develop semi-supervised and supervised systems. But want to know about state-of-art systems and potentially successful approaches.
First of all you have to develop Q&A Data in a sufficient manner. After collect the data design a system which will collect randomly Question from your database and see the accuracy of selecting the Question from the database. Note that the Q&A on the topic should be in group and it should selected randomly by the system. Get the accuracy of selecting the Q&A from the database.
You can get data from any cQA site i.e., yahoo, Stackoverflow, Qoura etc.
Especially, check the following site ,it might be indirectly helpful for getting sufficient amount of data as well as concepts and techniques.
http://alt.qcri.org/semeval2017/task3/
And for implementations, I would like to recommend deep models.
I think the following link will be a good start .
https://www.kaggle.com/c/quora-question-pairs
Thank you for answers!
I aim to obtain a simple dataset of questions from real life(avg 5-6 words per sentence) with a limited context. Because I am not a semanticist, baseline of the model would be limited to morphology/syntax interface but I will definitely move towards pragmatics. At the baseline I want to use a tagset for annotating questions and train a deep learning model(depending of the training data size).
Should I develop my own tagset or use standardized one? is the first of the questions.
Secondly, I have knowledge of Bayesian learning but I think this project will require more knowledge on statistical QA knowledge even n-ary systems may not be enough.
@Bahadorezza I would definitely like to read your research.
Dear Serkan,
My be this link is usefull for you
https://www.researchgate.net/publication/305271182_Pattern_Discovery_using_QG_for_Question-Answering_Pairs
Article Pattern Discovery using QG for Question-Answering Pairs
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Artificial Intelligence