graph - Algorithms for learning user inputs, and for offering suggestions -

i'm searching algorithm respectively method learning user actions (inputs) in program, and, based on built information base of done user actions, offer suggestions future actions user. information base should built actions of multiple users using same software. user actions dependent on order in occur. means, suggestion should made based on done user actions in session. session abstract time period, in user works software.

in initial approach, thought of moddeling user actions in directed graph, each node represents unique user action instance. user action, done first time, generates new node. nodes have counter representing how user did user action. transition 1 node exists, when user action done after 1 (modelling sequence of user actions). every transition, probability computed based on counters of subsequent nodes (i.e. nodes, there transition). there root node starting point, directs initial nodes (user actions done first in session). (hidden) markov model, not sure. not bayesian network, because can cyclic graph (desireable).

are there methods, algorithms, libraries etc. problem? if not, how approach? alternatives, better ideas?

it seems trying "recommendations", in case feature "actions" done users. considering both personalization (which builds model each user based on his/her behavior) or collaborative filtering (which builds model associatively finding similar users) general direction

what suggesting hmm in consider observation user action , state user intend. based on real application, might work or not.

another more promising way model directly build discriminative model (logistic regression, svm) using previous user action , user history feature predict next move. if have enough data, work not bad