You missed two points. Firstly "at different times" meaning people who are all there at 3pm one day, 10am a different day, and 4:30pm on another day. That's a strong indicator that they're not there coincidently if they don't appear there at other times as well. Secondly you missed where I said "likely to know each other". Mining large data sets for relationships is all about probability, as nothing is really guaranteed to be 100% certain, but you can get to quite high confidence levels with relatively small amounts of data.