This is a very bad idea, let me explain why...

##Story Time##

In the beginning people had questions and were looking for answers. So these people went to forums and mailing lists to find these answers from the experts. The system was imperfect though, some might say broken; questions were frequently repeated, it was difficult to know what the answer to a question was, and no one knew who to trust. Now moderators and Karma systems worked hard to correct these problems, but the people cried out: We want a better system!

So a man named Jeff teamed up with a man named Joel and they designed a new system. Where forums had been designed for people having a conversation, Jeff and Joel designed a system for Q&A. Drawing on their knowledge of of Q&A's, search engines and human behavior on the internet, the worked hard to create a system where people could find the answers to their questions. They emphasized good answers "floating," reputation and distinguishing between questions, comments and answers.

And all was good... until some people wanted to start using this software for something other than Q&A.

##The Problem## My endearing story aside, the StackExchange Engine is a very bad tool to use for things like: Recommendations and Polling (I'm ignoring comics and chatting for the moment). It suffers from a very large number of problems which have been enumerated before. A well designed recommendation system would profile characteristics and similarities and try to determine the relationships between them. This is very much at odds with a system designed to solve "needle in the hay stack" style problem. Instead of selecting the "right" answer, a recommendation wants to absorb "what" the relationship between A and B is, not the strength. While many people seem to think that the voting system will float the "best" recommendation, the reality is that it will float the "most popular" recommendation, not the one most related. What's worse, all those other recommendations don't gain the knowledge of "how" they are related, only the magnitude of that relationship.

If I were to design a recommendation engine based on crowd sourcing, I wouldn't index based on people looking for recommendations, I'd index based on discrete things (for example, in a video game recommendation engine I'd index on Video Games). From there I'd want users to assign characteristics that the games share: Metal Gear Solid is like Arkham Asylum because both are Stealth games. This way people don't see one and think they're both games about Batman. Finally, I wouldn't want to assign reputation to people based on agreement, but rather have a metric of how close one user's "taste" is to another user. Since the quality of a given person's taste isn't an objective value, similarity in taste is more important than correctness.

Ultimately the StackExchange Engine lacks the qualities of a good Recommendation Engine. It is designed to find the "correct" answer, a meaningless term when considering Recommendations have more than one correct answer and that they are correct in differing ways. Using SE for these things is like using Forums for Q&A, or using Microsoft Word to do a Spreadsheet: Just Plain Wrong.

So let's not repeat the same mistake our predecessors did, and try to abuse the engine into handling this content. I realize when you have a good hammer, how everything looks like nails, but let's try and use the right tool for the job here.

This is a very bad idea, let me explain why...

Story Time

In the beginning people had questions and were looking for answers. So these people went to forums and mailing lists to find these answers from the experts. The system was imperfect though, some might say broken; questions were frequently repeated, it was difficult to know what the answer to a question was, and no one knew who to trust. Now moderators and Karma systems worked hard to correct these problems, but the people cried out: We want a better system!

So a man named Jeff teamed up with a man named Joel and they designed a new system. Where forums had been designed for people having a conversation, Jeff and Joel designed a system for Q&A. Drawing on their knowledge of of Q&A's, search engines and human behavior on the internet, the worked hard to create a system where people could find the answers to their questions. They emphasized good answers "floating," reputation and distinguishing between questions, comments and answers.

And all was good... until some people wanted to start using this software for something other than Q&A.

The Problem

My endearing story aside, the StackExchange Engine is a very bad tool to use for things like: Recommendations and Polling (I'm ignoring comics and chatting for the moment). It suffers from a very large number of problems which have been enumerated before. A well designed recommendation system would profile characteristics and similarities and try to determine the relationships between them. This is very much at odds with a system designed to solve "needle in the hay stack" style problem. Instead of selecting the "right" answer, a recommendation wants to absorb "what" the relationship between A and B is, not the strength. While many people seem to think that the voting system will float the "best" recommendation, the reality is that it will float the "most popular" recommendation, not the one most related. What's worse, all those other recommendations don't gain the knowledge of "how" they are related, only the magnitude of that relationship.

If I were to design a recommendation engine based on crowd sourcing, I wouldn't index based on people looking for recommendations, I'd index based on discrete things (for example, in a video game recommendation engine I'd index on Video Games). From there I'd want users to assign characteristics that the games share: Metal Gear Solid is like Arkham Asylum because both are Stealth games. This way people don't see one and think they're both games about Batman. Finally, I wouldn't want to assign reputation to people based on agreement, but rather have a metric of how close one user's "taste" is to another user. Since the quality of a given person's taste isn't an objective value, similarity in taste is more important than correctness.

Ultimately the StackExchange Engine lacks the qualities of a good Recommendation Engine. It is designed to find the "correct" answer, a meaningless term when considering Recommendations have more than one correct answer and that they are correct in differing ways. Using SE for these things is like using Forums for Q&A, or using Microsoft Word to do a Spreadsheet: Just Plain Wrong.

So let's not repeat the same mistake our predecessors did, and try to abuse the engine into handling this content. I realize when you have a good hammer, how everything looks like nails, but let's try and use the right tool for the job here.

This is a very bad idea, let me explain why...

##Story Time##

In the beginning people had questions and were looking for answers. So these people went to forums and mailing lists to find these answers from the experts. The system was imperfect though, some might say broken; questions were frequently repeated, it was difficult to know what the answer to a question was, and no one knew who to trust. Now moderators and Karma systems worked hard to correct these problems, but the people cried out: We want a better system!

So a man named Jeff teamed up with a man named Joel and they designed a new system. Where forums had been designed for people having a conversation, Jeff and Joel designed a system for Q&A. Drawing on their knowledge of of Q&A's, search engines and human behavior on the internet, the worked hard to create a system where people could find the answers to their questions. They emphasized good answers "floating," reputation and distinguishing between questions, comments and answers.

And all was good... until some people wanted to start using this software for something other than Q&A.

##The Problem## My endearing story aside, the StackExchange Engine is a very bad tool to use for things like: Recommendations and Polling (I'm ignoring comics and chatting for the moment). It suffers from a very large number of problems which have been enumerated before. A well designed recommendation system would profile characteristics and similarities and try to determine the relationships between them. This is very much at odds with a system designed to solve "needle in the hay stack" style problem. Instead of selecting the "right" answer, a recommendation wants to absorb "what" the relationship between A and B is, not the strength. While many people seem to think that the voting system will float the "best" recommendation, the reality is that it will float the "most popular" recommendation, not the one most related. What's worse, all those other recommendations don't gain the knowledge of "how" they are related, only the magnitude of that relationship.

If I were to design a recommendation engine based on crowd sourcing, I wouldn't index based on people looking for recommendations, I'd index based on discrete things (for example, in a video game recommendation engine I'd index on Video Games). From there I'd want users to assign characteristics that the games share: Metal Gear Solid is like Arkham Asylum because both are Stealth games. This way people don't see one and think they're both games about Batman. Finally, I wouldn't want to assign reputation to people based on agreement, but rather have a metric of how close one user's "taste" is to another user. Since the quality of a given person's taste isn't an objective value, similarity in taste is more important than correctness.

Ultimately the StackExchange Engine lacks the qualities of a good Recommendation Engine. It is designed to find the "correct" answer, a meaningless term when considering Recommendations have more than one correct answer and that they are correct in differing ways. Using SE for these things is like using Forums for Q&A, or using Microsoft Word to do a Spreadsheet: Just Plain Wrong.

So let's not repeat the same mistake our predecessors did, and try to abuse the engine into handling this content. I realize when you have a good hammer, how everything looks like nails, but let's try and use the right tool for the job here.

This is a very bad idea, let me explain why...

##Story Time##

In the beginning people had questions and were looking for answers. So these people went to forums and mailing lists to find these answers from the experts. The system was imperfect though, some might say broken; questions were frequently repeated, it was difficult to know what the answer to a question was, and no one knew who to trust. Now moderators and Karma systems worked hard to correct these problems, but the people cried out: We want a better system!

So a man named Jeff teamed up with a man named Joel and they designed a new system. Where forums had been designed for people having a conversation, Jeff and Joel designed a system for Q&A. Drawing on their knowledge of of Q&A's, search engines and human behavior on the internet, the worked hard to create a system where people could find the answers to their questions. They emphasized good answers "floating," reputation and distinguishing between questions, comments and answers.

And all was good... until some people wanted to start using this software for something other than Q&A.

##The Problem## My endearing story aside, the StackExchange Engine is a very bad tool to use for things like: Recommendations and Polling (I'm ignoring comics and chatting for the moment). It suffers from a very large number of problems which have been enumerated before. A well designed recommendation system would profile characteristics and similarities and try to determine the relationships between them. This is very much at odds with a system designed to solve "needle in the hay stack" style problem. Instead of selecting the "right" answer, a recommendation wants to absorb "what" the relationship between A and B is, not the strength. While many people seem to think that the voting system will float the "best" recommendation, the reality is that it will float the "most popular" recommendation, not the one most related. What's worse, all those other recommendations don't gain the knowledge of "how" they are related, only the magnitude of that relationship.

If I were to design a recommendation engine based on crowd sourcing, I wouldn't index based on people looking for recommendations, I'd index based on discrete things (for example, in a video game recommendation engine I'd index on Video Games). From there I'd want users to assign characteristics that the games share: Metal Gear Solid is like Arkham Asylum because both are Stealth games. This way people don't see one and think they're both games about Batman. Finally, I wouldn't want to assign reputation to people based on agreement, but rather have a metric of how close one user's "taste" is to another user. Since the quality of a given person's taste isn't an objective value, similarity in taste is more important than correctness.

Ultimately the StackExchange Engine lacks the qualities of a good Recommendation Engine. It is designed to find the "correct" answer, a meaningless term when considering Recommendations have more than one correct answer and that they are correct in differing ways. Using SE for these things is like using Forums for Q&A, or using Microsoft Word to do a Spreadsheet: Just Plain Wrong.

So let's not repeat the same mistake our predecessors did, and try to abuse the engine into handling this content. I realize when you have a good hammer, how everything looks like nails, but let's try and use the right tool for the job here.