The invention discloses a user negative emotion prediction method and system based on web logs. The method comprises the steps that (1) web log data of a user is acquired; (2) active search data initiated by the user is acquired according to the web log data, and search term extraction is performed on the active search data; (3) word segmentation is performed on a search term to form multiple words divided according to properties; (4) semantic similarity calculation is performed on a word segmentation result and a preconfigured sensitive word; and (5) whether matching of the sensitive word succeeds is judged according to the degree of approximation of semantic similarity calculation, and the negative emotion of the user is predicted according to the judgment. Through the user negative emotion prediction method and system, a web dangerous speech, a negative trend and other states of the user can be discovered as early as possible according to the web log data of the user for intervention treatment.