Sounds like a good basis for neural networks. Only problem is you need a LOT of data to successfully make a chess engine and all known games of one particular player is probably not nearly enough. I do like the idea though and it does have some potential.
Somehow I feel someone else already had this idea before me, but it failed. The programmers of today are trying to make a perfect chess engine that could beat anyone no matter what. The thing is, engines don't have preferred moves, they just play whatever has the best chance to success, and if there are multiple moves that can do well equally, they pick a random one. They don't have their own opinion! But what if we can fake it... So, by analyzing games of a particular player, an engine could play preferred moves of that player in situations which can be answered equally good in multiple ways! And, of course, it could also play less preferred moves, but only as much as that player would in correspondence to the preferred moves! Take the World Champions, as they have the widest pallette of successful answers and some of the biggest archives of games, and voila - you could feel like playing against Morphy, Tal, Kasparov, Capablanca, Karpov, Fischer, Petrossian etc. To make the engine's games even more realistic, you could add handicaps which add some stubbornness to the engine, thus making it play worse not just by playing any slightly less effective move, but by playing a slightly less effective preferred move! I think it could also be a great way to learn, understanding the thought process of the chess legends and also learning to exploit any weaknesses they may have or have had, through a game of chess! What do you think?