Post: #126

(01-26-2014 05:57 AM)username Wrote:  

(01-25-2014 09:22 PM)gringochileno Wrote:  Anybody know how one might go about figuring out which match questions girls are answering?

On each girls profile it says which questions they answered.

It would be pretty simple to make scripts that:

1) Add to a database all the girls that match your criteria.
2) Go through the database one by one and extract all her answers as well as biography and other stuff from her profile page and put it the databases.
3) With the database of questions you can easily sort by the one with the most results and which answer is the most popular. Could also add additional information like race and body weight to get even more accurate results. (ex: latinas who are slim, athletic, or average answer question #234 yes 90% of the time.)

If I have some time next week after I get some shit I need to do done I will post some sample code. Actually I already have 20% of the code written in PHP (with MySQL) from previous data extraction research projects.

Awesome! Let me know if you do any more work on this. I have a math background and I could probably work out the dude's weighting scheme if we figure out how to get our hands on the data.

Post: #127

(01-26-2014 07:37 PM)gringochileno Wrote:  

(01-26-2014 05:57 AM)username Wrote:  

(01-25-2014 09:22 PM)gringochileno Wrote:  Anybody know how one might go about figuring out which match questions girls are answering?

On each girls profile it says which questions they answered.

It would be pretty simple to make scripts that:

1) Add to a database all the girls that match your criteria.
2) Go through the database one by one and extract all her answers as well as biography and other stuff from her profile page and put it the databases.
3) With the database of questions you can easily sort by the one with the most results and which answer is the most popular. Could also add additional information like race and body weight to get even more accurate results. (ex: latinas who are slim, athletic, or average answer question #234 yes 90% of the time.)

If I have some time next week after I get some shit I need to do done I will post some sample code. Actually I already have 20% of the code written in PHP (with MySQL) from previous data extraction research projects.

Awesome! Let me know if you do any more work on this. I have a math background and I could probably work out the dude's weighting scheme if we figure out how to get our hands on the data.

Once you have the data that's usually most of the work. It then becomes a classic segmentation analysis using survey data. I would heavily recommend statistical software for the sorting and classification of the data. The math dude was probably using python + scipy package or R since they're both free high quality data mining tools. Unfortunately, they're pretty code driven so if you want to avoid that you could use SPSS but you usually have to pay for that one (unless you pirate it).

From there you basically:

Format the data for a unsupervised learning algorithm .That means possibly performing a data transformation step transforming the data to between 0 or 1, -1 an 1, or standard deviations. Since the data is already in 0' and 1's you can probably skip this step. You then apply a dissimilarity/similarity measure. Most cases that's squared euclidean distance.

When running an unsupervised method, you have to make sure it finds the correct ordering and classification of the data. Since k-means/mode/median have a dependency on where the initial start point is, run the unsupervised learning method until the clustering results are stable, we usually do this about 1000 times. In the article he found the stable number of clusters is 7 so you probably need to do less guessing about the correct size and just put in 7 and let the method run to the optimal segmentation. Or you can do hierarchical clustering but that's pretty cpu/memory intensive.

Then you use a supervised learning method to predict the cluster by the questions and answered used. He used adaboost which I believe is basically a more advanced form of regression analysis with resampling/bootstrapping from the data set. You should be using a method that lets you see the importance weights so this could be regression, decision trees, and their resampling derivatives.

I might try reaching to the guy since we work in the same field and I went to one of his schools.

Post: #128

(01-26-2014 09:33 PM)cibo Wrote:  

(01-26-2014 07:37 PM)gringochileno Wrote:  

(01-26-2014 05:57 AM)username Wrote:  

(01-25-2014 09:22 PM)gringochileno Wrote:  Anybody know how one might go about figuring out which match questions girls are answering?

On each girls profile it says which questions they answered.

It would be pretty simple to make scripts that:

1) Add to a database all the girls that match your criteria.
2) Go through the database one by one and extract all her answers as well as biography and other stuff from her profile page and put it the databases.
3) With the database of questions you can easily sort by the one with the most results and which answer is the most popular. Could also add additional information like race and body weight to get even more accurate results. (ex: latinas who are slim, athletic, or average answer question #234 yes 90% of the time.)

If I have some time next week after I get some shit I need to do done I will post some sample code. Actually I already have 20% of the code written in PHP (with MySQL) from previous data extraction research projects.

Awesome! Let me know if you do any more work on this. I have a math background and I could probably work out the dude's weighting scheme if we figure out how to get our hands on the data.

Once you have the data that's usually most of the work. It then becomes a classic segmentation analysis using survey data. I would heavily recommend statistical software for the sorting and classification of the data. The math dude was probably using python + scipy package or R since they're both free high quality data mining tools. Unfortunately, they're pretty code driven so if you want to avoid that you could use SPSS but you usually have to pay for that one (unless you pirate it).

From there you basically:

Format the data for a unsupervised learning algorithm .That means possibly performing a data transformation step transforming the data to between 0 or 1, -1 an 1, or standard deviations. Since the data is already in 0' and 1's you can probably skip this step. You then apply a dissimilarity/similarity measure. Most cases that's squared euclidean distance.

When running an unsupervised method, you have to make sure it finds the correct ordering and classification of the data. Since k-means/mode/median have a dependency on where the initial start point is, run the unsupervised learning method until the clustering results are stable, we usually do this about 1000 times. In the article he found the stable number of clusters is 7 so you probably need to do less guessing about the correct size and just put in 7 and let the method run to the optimal segmentation. Or you can do hierarchical clustering but that's pretty cpu/memory intensive.

Then you use a supervised learning method to predict the cluster by the questions and answered used. He used adaboost which I believe is basically a more advanced form of regression analysis with resampling/bootstrapping from the data set. You should be using a method that lets you see the importance weights so this could be regression, decision trees, and their resampling derivatives.

I might try reaching to the guy since we work in the same field and I went to one of his schools.

Haha awesome, I love this forum. By the way I bought the guy's book (it's 5 bucks on Amazon) and he describes a low-tech method that seems to approximate what he actually did by scraping all the data and doing the k-modes analysis. I'm gonna try it out and see if it improves my match percentages or leads to more messages. PM me if you want the relevant section.

Post: #129

^^^ Alright so is there any way for someone who doesn't have a PhD in math to make any use of this stuff? Because I have no idea WTH you guys are even talking about up here.

Post: #130

(01-26-2014 11:09 PM)speakeasy Wrote:  ^^^ Alright so is there any way for someone who doesn't have a PhD in math to make any use of this stuff? Because I have no idea WTH you guys are even talking about up here.

I'm trying the low-tech approximation of his method that he describes in the book right now. He says you need to strategically answer 500 questions for best results and I'm through about 100, I'll report back if it makes a difference when I'm done.

edit: this had better fucking work. Answering match questions gets incredibly annoying after the first 50 or so.

Post: #131

So far this seems to be working. I have a lot more 90%+ matches than I had before despite having answered about the same number of questions. The real litmus test will be to see if this gets converted into messages though.

Post: #132

Have any women called him "creepy" yet for simplifying his quest to find a woman he finds very desireable?

Post: #133

Alright so this is basically the guy's strategy (Dr. McKinlay: if you are reading this and would like me to take it down, I will ask a moderator to do so.):

1. Optimize match percentage by answering the right questions and giving them the right importance level.

This consists of (a) figuring out which match questions the girls you want are answering and (b) Answering the ones with at least one answer choice that you're confident those girls will not select. He gives the upshot of his statistical analysis, which is that the following rules generally apply:

How to answer match questions

• Avoid "divisive" questions. Answer only questions with one or more "throwaway" answers--answers that you're very sure girls won't pick. Then--and this is the crucial part--you choose as "acceptable" answers everything except the throwaway answer. An example would be a question asking whether you torture animals. You know the girl's not going to say she tortures animals so you confidently mark every answer other than "I torture animals" as an "answer I'll accept." This means the girl will be matched with you on that question as long as she didn't pick "I torture animals."
• If there aren't any throwaway answers to a question, play it safe: check every box under "Answer I'll accept," which will automatically set the importance of the question you assign to "irrelevant." This is useful for answering questions you want to answer anyway, or questions where you know what the girl is going to accept but don't have an answer that almost nobody will pick.
• The importance to attach to a question is dictated by the number of non-throwaway answers (e.g., a 4-answer-choice question where one of the answers is "I torture animals" has 3 non-throwaway answers). Questions with 3 non-throwaway answers get "Mandatory," questions with 2 non-throwaway answers (usually 3 total answers) get "Very Important," and questions with 1 non-throwaway answer (2 total answers, e.g. a yes/no question) get "Somewhat Important."
• Answer questions in the order of importance to the girl. Go to the girl's match questions and sort them by "She cares about" to order them this way.

2. Answer 500 questions.

First, delete all your previous responses to questions.

Next, find questions to answer by going to the "Questions" section of your profile and start skipping through the questions, answering them as described above. Answer 20 questions in this way.

Now, go to "Matches" and search for girls who like guys in your desired age range within 100 miles of you, ordering the search results by match percentage. Pick one of the top girls (I usually pick the hottest one in the top 3) and go to their match questions ("The Two of Us"). Order those questions by "She cares about" and answer 10 questions in the same way. These are the girls you're going to optimize for so choose carefully at this stage.

Go back to the Matches section and do the exact same search again. Pick another girl and repeat the process with her match questions. Do this over and over until you've answered 500 questions, or you stop seeing new matches every time you search.

3. Seed your profile with common keywords.

The author mined the data for frequently-occurring words and topics. Imitate this by coming up with a list of likely keywords and do search for them individually, counting the number of 90+ matches that appear in the search results. Find places in your profile to strategically insert some of the highest-scoring words.

4. Get exposure

Allow your profile views to be public, and visit every profile of the girls you match. A script to automate this would be ideal, but you can also just do a basic local search, order it by match percentage, and ctrl-click on everything to open each profile a new tab. Each one of these clicks will show up as a profile view to the girl, and the idea is that a certain percentage of these girls will message you unsolicited after seeing you visited her profile. You can repeat this about once a week.

As an aside, I'm not sure whether this last step is advisable or not. I'm gonna stop after answering the match questions for now and see if that makes a difference, but the author claims he got about 1 message for every 10 profile views by taking this step.

I think that about covers it. The rest of the book is just him discussing what messages to send and what to do on dates, which you can safely ignore. Once you start getting messages rolling in just run your regular online game.

Post: #134

Well I had to kill a whole bottle of wine to make it through, but I answered 500 questions using the method described above last night. It definitely does improve your match percentages when you answer questions this way--I have many, many pages of 90%+ matches in my area now, whereas before I only had a few. However, I still haven't gotten a message or noticed an increase in profile views as a result (although I have myself set to invisible when I view other people--I wanted to only change one thing at a time to see if it makes a difference).

I'm gonna leave it like this for a few more days. If I still don't see an improvement in my results I'll try steps 3 and 4 where you find common keywords and fish for profile views by clicking on people.

Post: #135

Just got my first message as a result of doing the hack. Girl messaged me with "We have so much in common!" (97% match.) I'm gonna try and bang.

Post: #136

What questions did you answer that gave you a higher match%

Post: #137

(01-28-2014 03:25 PM)J.J. Wrote:  What questions did you answer that gave you a higher match%

Follow the procedure I posted above. You let the girls dictate which questions you answer by finding the ones they marked as important to them.

Post: #138

gringo would have repped you for that but WIA already did. Many thanks...not sure if I want to get into OKC, but awesome summarizing of the work.

Post: #139

(01-27-2014 01:30 AM)gringochileno Wrote:  Avoid "divisive" questions. Answer only questions with one or more "throwaway" answers--answers that you're very sure girls won't pick.

This one is harder than it seems. There's a lot of 2 answer questions, and while some are easy (torturing animals, etc), some are tricky. For example, I've seen a lot of chicks get the Which is Bigger? question wrong. So you have to ask yourself, am I going to be able to stand being around a chick who thinks the Earth is bigger than the sun? Also anything to do with smoking, politics, birth control, sports, drugs/alcohol, gender relations, rape (yes, there's a bunch of questions about rape fantasies, "does no really mean no?" etc), even video games - you're going to find people strongly opposed. So there's some guess work. I take on the video games thing because I like more active/athletic chicks and they usually don't like spending all day in front of the XBox.

Similarly, the "inverted glove" question, I see chicks get wrong and then mark me down for getting it right, or the "what does wherefore mean?" - chicks mark "where" and ding you for picking "why" (Shakespeare pro tip: when Juliet says "wherefore art thou Romeo, she's asking "why did you have to be a Montague?" Google can make you seem smart, ladies). So the logic q's can be a trap.

I think that you have to do some targeting - set your q's, wait 24 hours, and then try to refine them based on your target.

Post: #140

(01-28-2014 04:47 PM)RockHard Wrote:  

(01-27-2014 01:30 AM)gringochileno Wrote:  Avoid "divisive" questions. Answer only questions with one or more "throwaway" answers--answers that you're very sure girls won't pick.

This one is harder than it seems. There's a lot of 2 answer questions, and while some are easy (torturing animals, etc), some are tricky. For example, I've seen a lot of chicks get the Which is Bigger? question wrong. So you have to ask yourself, am I going to be able to stand being around a chick who thinks the Earth is bigger than the sun? Also anything to do with smoking, politics, birth control, sports, video games - you're going to find people strongly opposed. So there's some guess work. I take on the video games thing because I like more active/athletic chicks and they usually don't like spending all day in front of the XBox.

Similarly, the "inverted glove" question, I see chicks get wrong and then mark me down for getting it right, or the "what does wherefore mean?" - chicks mark "where" and ding you for picking "why" (Shakespeare pro tip: when Juliet says "wherefore art thou Romeo, she's asking "why did you have to be a Montague?" Google can make you seem smart, ladies). So the logic q's can be a trap.

I think that you have to do some targeting - set your q's, wait 24 hours, and then try to refine them based on your target.

Yeah, if you're not sure whether a question has a throwaway answer it's best to skip it entirely or mark it irrelevant. The yes/no questions aren't too critical anyway, the real money is marking mandatory on the 4-answer questions that have a throwaway (when you do that you'd better be damn sure it's actually a throwaway though, since you lose a lot of match points if she doesn't pick one of your acceptable answers on a mandatory question).

I've had some success with refining answers too, if you go to a girl's questions and click "unacceptable" you can see the ones that didn't match and find questions you answered that ended up being divisive. I managed to marginally increase my number of high matches by doing some of this today.

Post: #141

I don't know if you guys are minorities, but I've used the mandatory/very important/etc designation a lot. I use it on questions like the interracial marriage and child of another race questions. Its a good way to weed out the girls who won't touch a black guy. There's like 4-5 questions similar to that so its important that you use them to filter out girls who aren't interested in dating outside of their race.

Post: #142

(01-28-2014 05:26 PM)J.J. Wrote:  I don't know if you guys are minorities, but I've used the mandatory/very important/etc designation a lot. I use it on questions like the interracial marriage and child of another race questions. Its a good way to weed out the girls who won't touch a black guy. There's like 4-5 questions similar to that so its important that you use them to filter out girls who aren't interested in dating outside of their race.

I'm white, but it's a good point. There's other q's like that, like the "are you born the same gender you are now". Yes/Yes/Mandatory. Slam dunk.

Post: #143

I've never gotten messaged based on a high match score, even though I see high matches all the time (also by skipping inflammatory questions and sticking to safe lukewarm topics). Maybe the pool of women in my area is too small. From my experience, women with medium match percentages (60-80%) usually have a higher reply rate than those 81-100%. Whenever I message a girl that's "perfect", it's usually game over already.

Post: #144

(01-29-2014 02:26 AM)Handsome Creepy Eel Wrote:  I've never gotten messaged based on a high match score, even though I see high matches all the time (also by skipping inflammatory questions and sticking to safe lukewarm topics). Maybe the pool of women in my area is too small. From my experience, women with medium match percentages (60-80%) usually have a higher reply rate than those 81-100%. Whenever I message a girl that's "perfect", it's usually game over already.

That's the first message I've ever gotten based on match score. It's possible it's a coincidence, although I doubt it since I've gone from under 20 90%+ matches to over 300 in the last day and a half. I do think a large part of the dude's success came from the script he wrote to visit everyone's profile (plus being tall, white and not ugly certainly helps). I'm still experimenting with this, we'll see what happens.

Post: #145

Here's a good one:
"Did you join OkCupid just so you could find people to have sex with?"
Yes or No

Post: #146

I know this is a bit late, but I'm surprised no one caught this little gem of honesty:

Quote:Back when he was the pursuer, he’d swapped three to five messages to get a single date. Now he’d send just one reply. “You seem really cool. Want to meet?”

Textbook game theory.

Post: #147

(01-21-2014 11:34 PM)DavidZRH Wrote:  I can't believe there are people here rating her a 4. WTF? She's a goddam 1 or 2.

Does being thin with no tits and no ass make her a 4?

Same thing I was thinking, makeup or no makeup I do not find this chick attractive at all nor would I be open to having sex with her even if she somehow blossomed into a 6.

Post: #148

(02-04-2014 04:45 AM)Distant Light Wrote:  

(01-21-2014 11:34 PM)DavidZRH Wrote:  I can't believe there are people here rating her a 4. WTF? She's a goddam 1 or 2.

Does being thin with no tits and no ass make her a 4?

Same thing I was thinking, makeup or no makeup I do not find this chick attractive at all nor would I be open to having sex with her even if she somehow blossomed into a 6.

I think guys around here have a standard scale. 1 is a grotesque facial disfiguration, 2 is humorously unattractive. 3 includes the morbidly obese, "cute if she lost 200 lbs". 4 is just plain unattractive.

Someone else on here said 1-10 is bs, and all that matters is the boner scale, which I agree with. Personally, I think we should be using irrational numbers as the scale. These are women, after all.

Post: #149

I have way more profile views since optimizing my match questions, but nobody else messaging me outright. Even this is a positive if you like to message girls who view you but don't leave a message. I might try the mass-profile-visiting trick in a few days just to see if that only works for tall good-looking white guys.

Post: #150

Based on the ROK article anything that fails the boner test is below a six