According to one embodiment, an action history search device receives an inquiry from a user and outputs a inquiry time and a target of inquiry, decides a range for searching action history information representing history of the user's action together with a time of the user's action using the target of inquiry, and calculates an elapsed time from the time of the user's action within the range to the inquiry time. The device judges using the elapsed time and a narrowing-down model a probability on each response candidate to the inquiry based on the history of the action within the range. The narrowing-down model is used for judging according to the elapsed time the probability that the response candidate to the inquiry obtained from the history of the action is the user's desired response. The device outputs the response candidate according to the probability.