How do AI Systems gather information?
How you have wondered about how do recommendation systems read signals r intent? How do they accumulate data to recommend things that we inevitably like?
Mentioned below are a few common methods:
- Click based
In this case, systems accumulate data by reading the clicks as a favorable or acceptable signal. More the number of clicks, more is the probability that the person would like it.
Advantage of this method is that it has abundant data. We all tend to click to explore things (whether we purchase or watch that item or not, is not a concern here).
The downside of this method is that it is often not reliable. Some of us accidentally click on buttons, for instance - closing an ad; or are lured into clicking an ad like click here to know win a prize. So on a qualitative level, this might not be the best method. In today’s time, a lot of bots also operate that are made with a purpose of show fraudulent traffic. So a qualitative issue comes up here.
2. Purchase based
To overcome the quality issues of click based data accumulation, came the Purchase based behaviour where data would be captured and taken into account only if the user has purchased an item. If the click happened by mistake then s/he would not go ahead and purchase. And since no purchase happened hence it would not be taken into account.
The advantage of this system is that it is more fraud resistant to that of a click based system. And someone will have to buy a lot of items to skew the numbers. That is how big companies like Amazon have so much of data to work with for recommendations.
3. Consumption behaviour based
Not everywhere will it be about items you purchase, but consume. For instance YouTube. In that case how you consume the content, will be the data that will be accumulated and used for recommendation. In YouTube’s case, time watched can be treated as measure of how much you like. It is an indicator of consumption of time; so acts as a better gauge of interest.