[REF] ICS User Experience Graphs (sentiment analysis)

Oodie · Apr 21, 2012

bedalus said:
It's the code boxes I think. I think I'll upload individual Excel documents for each analysis.

Hopefully the pointy arrow is easily understood!

Yeah ! Pointy arrow is excellent.
Just messing with you

Sent from my Nexus S

tchaari · Apr 21, 2012

bedalus said:
Yes triples. Let's be practical here, there is no way that I can go through every post and input it into a box on a webpage, then collect the output and tally up the scores after every post for every thread.

It's that hard bedalus. A program can send a HTTP request and get the HTML that you can analyze as a string. I can do it in Java but I haven't any free time this period

bedalus said:
The link for the first one is down now. I didn't see any individual word colour coding when I tried it before. But it would take too much time to personally count up positives and negatives. I could read all the threads faster than doing that.

My idea is to count it using a program. These are the main steps:
1) Collect a limited number of random posts from each thread (without splitting or counting words)
2) Merge the posts in some big strings. Each string should have the length of the maximum characters supported by the sentiment analysis input box.
3) send each string in a HTTP request to the sentiment analysis web server.
4) get the HTML response as a string and automatically count the number of positives and negatives (if you want to use my first link).
4') If you prefer the second online tool (that outputs the arrow image), you can also get the resulting image automatically by the program.
If you want to do it in Java, I can give you more details on how to do it

. It should take one day of work at max

EDIT: first link is broken as you said bedalus. I'll try to look for other online opinion/sentiment analysis tools later.

tchaari · Apr 21, 2012

These other very interesting sentiment analysis online tools :
http://www.mashape.com/apis/Skyttle+API
http://text-processing.com/docs/sentiment.html
http://www.wingify.com/labs.php

glennkaonang · Apr 21, 2012

I think bedalus & tchaari should get the title "Recognized Analyst" here

Sent from my Nexus S using xda premium

tchaari · Apr 21, 2012

glennkaonang said:
I think bedalus & tchaari should get the title "Recognized Analyst" here

Sent from my Nexus S using xda premium

You are also in the same boat Glenn. The ICS ROM hierarchy is the proof

Bedalus is doing the hard work here. I just try to help by some ideas. I hope I had more free time to implement these ideas (that aren't always interesting lol)

glennkaonang · Apr 21, 2012

tchaari said:
You are also in the same boat Glenn. The ICS ROM hierarchy is the proof

That's an amateur work using Google Docs spreadsheet.
Even my nine-year-old nephew could do it

I was just doing that because I saw some people get it wrong when it comes to ROM sources.

Sent from my Nexus S using xda premium

bedalus · Apr 22, 2012

[AOSP][4.0.4]Codename Android 2.0.0RC|Fully Open Source+Nightly Builds
http://xdaforums.com/showthread.php?t=1463437

http://sentimentanalyzer.appspot.com/

bedalus · Apr 22, 2012

[ROM] [IMM76D] Pixel Rom 1.72 Pure & Simple Dexedrine Addicition
http://xdaforums.com/showthread.php?t=1411734

http://sentimentanalyzer.appspot.com/

bedalus · Apr 22, 2012

[ROM][MIUI 4][2.4.6] Catalyst MIUI (r2) with Multi-Language Pack
http://xdaforums.com/showthread.php?t=1483033

http://sentimentanalyzer.appspot.com/

Although I also checked the text here: http://text-processing.com/demo/sentiment/

and got a rating of 60% positive sentiment.

I think I will go back and check previous results with this service.

EDIT: Okay, I went back and checked Pixel ROM and got 80% negative!

So with these two websites, Catalyst's results are quite contradictory, and Pixel ROM's results are almost totally contradictory.

I suppose that this project has some serious technical difficulties to overcome if the results are going to be reliable.

tchaari · Apr 22, 2012

bedalus said:
EDIT: Okay, I went back and checked Pixel ROM and got 80% negative!

So with these two websites, Catalyst's results are quite contradictory, and Pixel ROM's results are almost totally contradictory.

I suppose that this project has some serious technical difficulties to overcome if the results are going to be reliable.

I think that this is mainly due to applying the sentiment analysis on words triples (which are not very significant). It should be better to analyze all the thread text. the wingify website (http://www.wingify.com/labs.php) can directly handle an URL and gives an overall score. You can continue to use the printable version of the forum to get a list of URLs and compute the average of pages scores for each thread. This is an example for the Catalyst thread:
http://xdaforums.com/printthread.php?t=1483033&pp=50&page=1
http://xdaforums.com/printthread.php?t=1483033&pp=50&page=2
...
http://xdaforums.com/printthread.php?t=1483033&pp=50&page=24

If the process is too long, you can take a limited number of pages randomly and just compute their average.
EDIT: we can consider the last 5 or 6 pages (200~300 posts). They are generally the most significant.
For example, the last 5 pages of the Catalyst thread are 75% positive (average of the 5 pages scores computed manually using http://www.wingify.com/labs.php)

bedalus · Apr 22, 2012

Nice work! I wonder if this can be automated? Also, perhaps sentiment can be tracked over time by supplying a set of posts from week to week.

kernels ; battery ; ROM ; gov/sched
Now with summaries in the first posts. Convenient for XDA app users!

bedalus · Apr 22, 2012

[ROM][4.0.4]Slim ICS 68MB- Clean, Simple & Fast - The Google Way
http://xdaforums.com/showthread.php?t=1390118

http://sentimentanalyzer.appspot.com/

And: http://text-processing.com/demo/sentiment/ gives 70% negative

EDIT: Again, this sentiment analysis stuff is a work in progress. tchaari is helping to improve the accuracy, but we may need to change the way we mine the data from the threads.

bedalus · Apr 22, 2012

Found this: http://wiki.metalayer.com/index.php?title=API_Sentiment

I can do this in windows:

curl http://api.metalayer.com/s/datalayer/1/sentiment -d "text=In 1972, a crack commando unit was sent to prison by ... A-Team."

Returns a slightly negative sentiment. (-0.5 on a scale from -5 to 5)

...using this version of curl: http://www.paehl.com/open_source/?CURL_7.24.0
(top most download)

I also found this list of sites: https://sites.google.com/site/twittersentimenthelp/other-resources

bedalus · Apr 22, 2012

These people seem to have a nice service (including a free option): http://www.alchemyapi.com/api/gstarted.html

...and there is an SDK so I may try to integrate this directly into my C++ program rather than relying on curl.

tchaari · Apr 22, 2012

bedalus said:
Nice work! I wonder if this can be automated? Also, perhaps sentiment can be tracked over time by supplying a set of posts from week to week.

Yes bedalus. It can be automated. A HTTP request (call of a web page from a program with strings at the place of input box) can be done using any language curl, C++, Java...
This is discussion about how to do it in C++
http://stackoverflow.com/questions/1011339/how-do-you-make-a-http-request-with-c

bedalus · Apr 22, 2012

tchaari said:
Yes bedalus. It can be automated. A HTTP request (call of a web page from a program with strings at the place of input box) can be done using any language curl, C++, Java...
This is discussion about how to do it in C++
http://stackoverflow.com/questions/1011339/how-do-you-make-a-http-request-with-c

Wow, programming is amazing. I'll try to build that example into a c++ program tomorrow. Then I'll see if I can do a full page of text to the api (there is a character limit). I'll aim to attach dates too.

http://www.alchemyapi.com/api/sentiment/textc.html

The above link gave me a slightly positive rating for my rank of word-triples for Slim ICS, whereas other sites had given negative. After looking at the word triples manually, it seems the sentiment is definitely positive. I think this is a good sign that the api will prove to be reliable... perhaps it will be more reliable with virgin text. Exciting!

bedalus · Apr 23, 2012

UPDATE: Checked my original program for tallying threads worked when threads got longer. Expected behaviour was to only read the necessary pages and update the .dat file. Unfortunately, it reads the new pages, and replaces the .dat file. Fixed now (I think... I have to wait until a thread gets longer to test).

I cannot understand how to integrate curl into c++ so I will just use system calls and place a prebuilt curl program in system32. It's an easy workaround.

bedalus · Apr 23, 2012

UPDATE: Nearly have it working.

I have threads sending their entire text (but no quotes) to the sentiment api in 400 word chunks, with scores returned.

I plan to share the results as a graph.

tchaari · Apr 24, 2012

bedalus said:
UPDATE: Nearly have it working.

I have threads sending their entire text (but no quotes) to the sentiment api in 400 word chunks, with scores returned.

I plan to share the results as a graph.

wow, this is very very interesting bedalus. Are you interested to mount a software engineering company with me? lol
EDIT: I am thinking about the next challenge: smoothness benchmark

bedalus · Apr 24, 2012

tchaari said:
wow, this is very very interesting bedalus. Are you interested to mount a software engineering company with me? lol
EDIT: I am thinking about the next challenge: smoothness benchmark

Don't know if I'll have the time to go global...

Program now works completely. I can't share it yet, I just need to write a method to import my private key from a file. Then other people can use it (but would need to get their own key). It wouldn't be much work, but unless I get demand for that, I actually think will just keep it private for now.

... because now it's working well, I will start working on a public chart!

EDIT: Okay, here's the new graph format, shows the accumulation of sentiment. So far only NexusMV (CM9 base) visible.

Link: https://docs.google.com/spreadsheet/ccc?key=0AntDDKv-lS6IdG9YU2h0LWVKT0RrZzVlMmZzNlhwSWc

[REF] ICS User Experience Graphs (sentiment analysis)

Which threads are the most useful to analyse?

ROM threads

Kernel threads

Theme/App threads

General threads

Threads for other devices

Inactive Recognized Developer

Senior Member

Senior Member

Senior Member

Senior Member

Senior Member

bedalus

Guest

Attachments

bedalus

Guest

Attachments

bedalus

Guest

Attachments

Senior Member

bedalus

Guest

bedalus

Guest

Attachments

bedalus

Guest

bedalus

Guest

Senior Member

bedalus

Guest

bedalus

Guest

bedalus

Guest

Senior Member

bedalus

Guest

Similar threads

Top Liked Posts