Brief:




I have been mulling on the idea of using neural nets in Human Resources in the area of resume parsing by approaching it from an angle of (text) pattern recognition ...

Comments welcome.

you would have to think about the depth of the system you want. Would it only be parsing at the syntax level, or do you want to deal with semantics as well?

And what are the features or patterns that you are looking for? Just key words?

Theres quite a lot to think about as they are many different approaches to the problem.

Yes, it is interesting. I was planning to do my research on my Final Year Project and from what I found out the following links are using ANI:-

http://eprints.nottingham.ac.uk/995/

http://eprints.nottingham.ac.uk/621/

Because of the fact that intrusion detection systems work based on pattern based recognitions, by using a simulated human immune system which is self learning using neural networks, the barriers of ID can now be break through.

in the inital stage, i would like to go by keywords, where the net identifies some simple items like
Name
Address
Phone Number
Date of Birth

and tags them accordingly.

if they tag incorrectly i will have to reclassify the item to correct tag.

so that next time they can tag it properly.

I am still thinking about how to teach the net...


Urgh more reading to be done

at this point i also not sure how potential viruses are detected , i belief each AV vendor has their own pattern recognition heuristics involve in this...

Not all. I know ESET implements heuristics. I'm not sure about others. Others are most likely polymorphic detection engines which might not even depend on heuristics at all.

It's like logic branching if this part of the code resembles this signature, and the other one, flag as possible threat or some sort. That's why there are false positives too.

I've passed your question to my friend who does natural language processing.

BTW if you're looking at parsing something else, our garage sales topics could use some machine categorization I would be happy to give you the data you need.

What is the final aim of the system?

Do you want to bin the resumes into seperate groups depending on what key words exist? Or something else?

You'd also have to first build a parser that will be able to read the resume input, and parse it into a format suitable for training the NN.

ah thats the keyword, why didnt it come across to me earlier...
for all i know i may be barking up the wrong tree on this

Well I have this project sitting on the shelf for many months, its to do with taking resumes of all formats (text readable ones)
and extract the useful data. As we know , many people write them in differing styles, but the general idea is

Personal Info
Education
Work Experience
Summaries (in paragraphs)

So i'd like to
1) split the data up to these categories so that its more helpful to my search
2) categorize the resumes by role (system engineer, programmer, analyst)

I'd probably focus on the IT resume sector at this moment...

Some months back it did occur to me that AI might somewhat help my efforts... I did do some of ANN pattern recognition yonks ago , probably around the time mariah carey released her first album


* my first attempt bombed miserably, which purely searched for keywords and marking entire blocks of text based on that...

This post has been edited by ahjames: Jun 17 2009, 08:58 AM

it's possible.. my Final Year Project was "Text Mining on resume" and i got A+ for my FYP... it was a research paper and i think mah univ did something to it liao... but its possible.. i created a prototype and demo-ed in front of the dean and some professors from other univ... and we ask some IBM HR ppl come and insert a training profile to be analysed by the data mining algorithm.. then some 'actual resume' that goes into a text parser.. and the result.. VOILA.. it matches the actual scenario...

---

Added on June 20, 2009, 3:29 pmehh.. just realized u're talking about neural networks... but it surely iz possible if combine neural networks + expert system with data mining to collect data though

This post has been edited by chezzball: Jun 20 2009, 03:29 PM

It depends on what you wanted to do. If you wanted to classify the resume into the appropriate roles, I do not think that ANN is the best choice, in the NLP world, methods like LDA or the more recent approach, HTMM are used to "learn" the few sample resumes that you created specifically for each of the roles. LDA or HTMM will basically take the words as a set of observations and assume that they can be explained by a set of unobserved groups, which in this case would be your company roles. The number of roles you can have have to be fixed. So if you have Software Engineer and System Analyst, you can expect your output says SE 0.8, SA 0.2 which is to say that the "model" believes that the resume is more suited to be an SE over an SA.

This post has been edited by zybler: Jul 12 2009, 01:16 AM

neural agent is one of the subject taugh during my univ period..
during my final year i was to develop a neural agent for driving simulation..it is easy actually however, need to be clear of the algorithm which is pain in the ass since i am no mathematician..

to simplify..the car itself has sensor, to detect speed, collision, obstacle and motion for every angle..the input for the neural network, then the algorithm will process the output such as speed and direction..i think can find the algo in some journal but cant remember which one..but for the project, im just using computer simulation that generate output, work well, but for the real project, we need real car and sensor and some mechanical part that moves the car..illustration:




its not so hard actually, this can be build using neuroD..i asked my lect who is a lecturer to help me on the algo..the final output will based on the input and the algorithm. for example, the car is moving at 70kmph, it detect other moving car at 50kmph 10 meter in front of it, then the the computer calculate and output is produce where the car will decelerate and slows at 50kmph before hitting the the car in front in 5 meter..same goes when entering corner, the car will detect the angle of the turning..it will process the input and generate output equivalent to the situation..btw, real AI car has been developed, i think the concept is the same as above..

This post has been edited by tentenko: Jul 14 2009, 10:30 AM

A Deep Artificial Neural Network is trained by Tom Lahore to generate and image of the Mona Lisa pixel by pixel. In order to generate each of the images, which when full size are 429-by-300 pixels, the neural network must be run 128,700 times.



And here is the end result:



Animated generation of Mona Lisa: