I am a pretty newbie in semantics, I have already asked a few questions here but still I am not clear about what to do.I am actually doing a university project whose end point is a Question Answer system based on semantics, need some help please. I have very short time as my time for the project is exhausting.

My understanding is that I have to convert all my data into ontologies first. Then I have to develop an front interface for the user to answer questions and connect the front interface to the ontology.I am not sure whether the questions should also be converted into triples or not?

I am following the following steps in order to achieve my goal:

Step 1:

I parse the plain text that I have using stanford parser

For Example:

I have a piece of plain text e.g;

"In this tutorial, I will show you the steps I took to create this Rough Woody Text Effect Made of Wood Splinters in Photoshop."

So this is what I have after parsing (using stanford online parser):

prep(show-7, In-1) det(tutorial-3, this-2) pobj(In-1, tutorial-3) nsubj(show-7, I-5) aux(show-7, will-6) root(ROOT-0, show-7) iobj(show-7, you-8) det(steps-10, the-9) dobj(show-7, steps-10) nsubj(took-12, I-11) rcmod(steps-10, took-12) aux(create-14, to-13) xcomp(took-12, create-14) det(Effect-19, this-15) nn(Effect-19, Rough-16) nn(Effect-19, Woody-17) nn(Effect-19, Text-18) dobj(create-14, Effect-19) partmod(Effect-19, Made-20) prep(Made-20, of-21) nn(Splinters-23, Wood-22) pobj(of-21, Splinters-23) prep(Splinters-23, in-24) pobj(in-24, Photoshop-25)

Someone here said that I already have the triples. So I assume I already have the triples.

Question 1: Please tell me if my understanding is correct?

Step 2:

I want to create the ontology (knowledge representation) for the parsed text I have

I assume that knowledge representation can be done by making ontologies for the above sentence.The problem is that I have large amount of data that needs to be converted into ontologies. It is impossible for me make ontologies manually. I need a automatic tool that converts my plain or parsed text to ontologies. I was suggested to use FRED by someone here but it seems that the ontologies created by FRED give errors when loading into ontology editors such as Protege 4.1, Top Braid Composer.

Question 2: Are there any GUI tools available so that I can convert my plain or parsed text automatically into ontologies?

Question 3: Please let me know whether I am going in the right direction. Is it necessary to convert triples into an ontology for the purpose of question answer system or can i directly run queries on triples. I am too confused by different ways and opinions of people. Also, what should be my next steps towards development of the QA system.Any suggestions/recommendations, please?

Waiting for your replies.Thank You All

asked 19 Jan '13, 20:17

sww's gravatar image

sww
4526
accept rate: 0%

edited 20 Jan '13, 01:47

@Sweet Burlap @fadirra @Signified I am really thankful for previous help, can you help further please?

(19 Jan '13, 20:22) sww sww's gravatar image

@harschware Can you help further please?

(19 Jan '13, 22:07) sww sww's gravatar image

So this is what I have after parsing (using stanford online parser): ...

Do you understand what that output is? I'm no expert on NLP, but it seems your parser has just Part-of-speech (POS)-tagged the sentence (EDIT: more specifically, building a dependency graph between words in the sentence; thanks @AKSWMember). This means that the machine has a rough idea of what the nouns and adjectives and prepositions, etc., are in the sentence.

That is something the NLP folks call "Shallow Parsing", which means that the machine has not really begun to analyse the semantics (i.e., meaning) of that sentence. Hence, from what you have parsed, all you can formally represent (e.g., in an ontology or even in RDF) is the structure of the sentence. Now you can match user questions and input sentences on a shallow level, but that will not buy you much by way of accuracy.

I want to create the ontology (knowledge representation) for the parsed text I have

As a question for you to think about, do you know what kind of ontology you want to create from your example sentence "In this tutorial, I will show you the steps I took to create this Rough Woody Text Effect Made of Wood Splinters in Photoshop."? In other words ... Could you do manually for that example what you want to program machines to do automatically for all examples? If not, I guess you're in trouble.

Because, if you want to automatically build a true machine-readable ontology representing the meaning of that sentence and sentences like it, you will have solved an AI-complete problem, namely natural language understanding, and I hope you will mention us in your Nobel prize acceptance speech. :)

In general, I strongly reiterate my previous advice that you should talk to an advisor or supervisor about your topic. It is my impression that your project is far too ambitious, esp. given the time constraints you have. Of course, I hope I'm wrong. :)

Please let me know whether I am going in the right direction. Is it necessary to convert triples into an ontology for the purpose of question answer system or can i directly run queries on triples.

It is not necessary to use triples or ontologies to build a QA-system. There has been 50 years of research into Question Answering, and most of the techniques heavily rely on NLP advancements.

permanent link

answered 21 Jan '13, 11:09

Signified's gravatar image

Signified ♦
23.9k1623
accept rate: 37%

edited 21 Jan '13, 17:46

1

The output is no simple POS tagging, but the generation of a dependency graph, which can shortly be described as a representation of grammatical relations between words in a sentence. This is somehow one step beyond the POS tagging. And yes, the thread owner should think about the meaning of this output first.

(21 Jan '13, 11:18) AKSWMember AKSWMember's gravatar image

Thanks. :) As I say, I'm no NLP expert.

(21 Jan '13, 11:19) Signified ♦ Signified's gravatar image

@Signified I am really thankful to you sir.

Though I am a newbie in both the fields whether it be semantics or QA systems, but I even feel the scope is too large.

The sentence I mentioned in the question was just to show the output. My actual work is to develop an QA system for Al-Quran based on semantic web. So can you suggest something on that area,please. I mean what should I focus on, what steps should I follow, etc

(21 Jan '13, 23:25) sww sww's gravatar image

Sorry no time for a long answer, but, I wonder if you might have better luck working with Open Calais ( especially since you are familiary with TBC). See this blog

I would suggest just getting some triples from Calais, and write SPARQL queries over that and see if you can "answer questions" without the need of a custom ontology first. Just get to know the Calais output and SPARQL, and then take your next steps. Then, post more questions in a day or two.

permanent link

answered 21 Jan '13, 00:19

harschware's gravatar image

harschware ♦
7.7k1616
accept rate: 20%

@harschware Thank you first of all. But Open Calais doesn't work with the text I have. I tried to use the online demo of Open Calais but it gives no results when I put in my text. And also I use Protege, I haven't used TBC really that much

Actually I have to develop an Question Answer system for a particular domain based on semantic web. The ontology is not available yet. (I wrote this just to give an idea of what I am aiming at)

Please suggest something else.

(21 Jan '13, 03:22) sww sww's gravatar image

You could also try Zemanta (similar to Calais)

(21 Jan '13, 11:28) harschware ♦ harschware's gravatar image

Question answering is no trivial problem, also not in the Semantic Web world. Have you ever tried to read some related work? I also worked on this topic, and the general approach was to process some natural language question and generate a SPARQL query based on this question.

permanent link

answered 21 Jan '13, 11:14

AKSWMember's gravatar image

AKSWMember
1.4k38
accept rate: 25%

@AKSWMember

I am thankful to you for your suggestions

I have gone through some literature but I am unable to derive something out of it. I am having real time constraints. My situation is that:

My actual work is to develop an QA system for Al-Quran based on semantic web. So can you suggest something on that area,please. I mean what should I focus on, what steps should I follow, etc

(21 Jan '13, 23:27) sww sww's gravatar image

Sorry, I do not understand how I can help you. There are many papers and one solution would be to 'transform' the question into a SPARQL query. But as I said, this is an open research topic. I guess nobody here can help you in creating a QA system, neither regarding the accuracy, nor in a short time. You should ask specific questions in this forum, question like 'How to create a QA system for domain X?' are at least for me out of scope.

(22 Jan '13, 16:09) AKSWMember AKSWMember's gravatar image

@AKSWMember Sir, I am just looking at the structure of the sentence and then base the question answer system based on that. Which means I look at the subject, verb, object, etc to form an triple and then I want to store them in Allelgerograph and run SPARQl on them. I want the output results somewhat like this sub(Verb,Object),etc. Can you suggest regarding this. And also how can i take this output and put into allegrograph to run SPARQL based Question Answers, i mean SPARQL Queries.

(28 Jan '13, 01:20) sww sww's gravatar image

This is not so easy as you think. The problem is to find the relevant information in a sentence. The extraction of RDF from unstructured data is still research topic. You can process a sentence with the Stanford dependency parser, but then the next step would be to decide what information is useful.

(28 Jan '13, 15:40) AKSWMember AKSWMember's gravatar image
Your answer
toggle preview

Follow this question

By Email:

Once you sign in you will be able to subscribe for any updates here

By RSS:

Answers

Answers and Comments

Markdown Basics

  • *italic* or _italic_
  • **bold** or __bold__
  • link:[text](http://url.com/ "title")
  • image?![alt text](/path/img.jpg "title")
  • numbered list: 1. Foo 2. Bar
  • to add a line break simply add two spaces to where you would like the new line to be.
  • basic HTML tags are also supported

Question tags:

×661
×30
×15
×11
×11

question asked: 19 Jan '13, 20:17

question was seen: 1,461 times

last updated: 28 Jan '13, 15:40