There is disclosed a method and system for generating a training set for training a machine learning algorithm (MLA) implemented in an information retrieval system. The method is executable by the server and comprises: retrieving, from a search log database of the server, a first query previously submitted to the server, a first SERP associated with the first query, a second query different from the first query and submitted after the first query, and a second SERP associated with the second query, the first query and the second query having been submitted by the electronic device: the first SERP including a first set of search results; and the second SERP including a second set of search results; in response to the second query being submitted within a same search session as the first query, generating the training set to be used as negative training examples for training the MLA.