# How to solve question answer questions using trees and graph theory?

Good evening! I'm doing an internship at Ecole Normale Supérieure of France, where they create their scholars, and I encountered a dataset I wanted to solve using graphs...

# I. The challenge ? Transforming Question Answering to tree parsing

"Quelle est la question ? Telle est la question !"

Question answering is not very funny for teachers, they are not encyclopedia, why not using graph theory to answer questions from paragraphs? That's why I want to try a tree attempt for a question answering dataset. I have questions and I have to find answers, or at least the paragraph that contains it in paragraphs with several sentences.

I want to know to what extent a tree representing a question $q_{ij}$ is included in another tree representing a paragraph $p_i$. A node represents a word and may include its synonyms. So I did an algorithm that seems not perfect yet and wanted to have your thoughts about it.

### input :

• $q_{ij}$ : When did Beyonce start becoming popular? in tree form :

Hopefully let say that popular as fame as a synonym.

.

          start
__________|___________
|    |     |     |  becoming
|    |     |     |     |
When did Beyonce  ?  popular

• $p_i$ : born and raised in houston, texas, she performed in various singing and dancing competitions as a child, and rose to fame in the late 1990s as lead singer of r&b girl-group destiny's child. in tree form

.

                                                      performed
________________________________________________________|__________________________________________________
|   |   |   |   |        |                               |                         |                       rose
|   |   |   |   |        |                               |                         |     ___________________|_________
|   |   |   |   |        |                               |                         |    |         |                   as
|   |   |   |   |        |                               |                         |    |         |                   |
|   |   |   |   |        |                               |                         |    |         |                 singer
|   |   |   |   |        |                               |                         |    |         |          _________|_____
|   |   |   |   |       born                             |                         |    |         |         |               of
|   |   |   |   |    ____|_____                          |                         |    |         |         |     __________|_____
|   |   |   |   |   |        raised                      in                        |    |         |         |    |              child
|   |   |   |   |   |     _____|_______                  |                         |    |         |         |    |                |
|   |   |   |   |   |    |             in           competitions                   as   |         in        |    |             destiny
|   |   |   |   |   |    |             |        _________|__________               |    |         |         |    |     ___________|______
|   |   |   |   |   |    |          houston    |                 singing         child  to      1990s       |    |    |                group
|   |   |   |   |   |    |             |       |          __________|_______       |    |     ____|____     |    |    |            ______|____
,  she  ,  and  .  and texas           ,    various     and              dancing   a   fame the       late lead r&b   's         girl


### Expected output:

series_0: rose to fame # from start becoming popular


or, even better rose to fame in the late 1990s

# II. My attempt ? This first algorithm :

i = 0
for node in q_ij:
if node is included in p_i
if node.father in series_i or series_i == Null:
series_i append node
else
i+=1
series_i append node
elif he is not we look n.synonyms is included
if node.father in series_i or series_i == Null:
series_i append node.synonyms
else
i+=1
series_i append node.synonyms

• Do you have a specific question? "I wanted to have your thoughts about it" seems too open-ended to be a good fit here (see our help center). It sounds like you want us to brainstorm approaches to your research problem. I'm not sure that is what this site is good at. – D.W. Aug 30 '18 at 23:06