Semantic item embedding vector training method and device

The invention discloses a semantic item embedding vector training method, and particularly, a first search word can be obtained, the first search word comprises at least two semantic items, any one ofthe at least two semantic items is called as a first semantic item, and the first semantic item has encyclopedic entries for convenient description. According to the invention, when the embedded vector of the first meaning item is trained, the keyword related to the first meaning item can be determined on the basis of the webpage related to the first meaning item and the encyclopedic entry of thefirst meaning item, and the embedded vector of the first meaning item is obtained through training by further utilizing the keyword related to the first meaning item. Therefore, according to the invention, when the embedded vector of the first meaning item is trained, the encyclopedic entry of the first meaning item and the webpage related to the first meaning item are considered, so that the accuracy of the embedded vect