acquiring a large amount of original ill text information in a network platform, and extracting entities of the original ill text information to obtain a plurality of ill word entities;

performing entity conversion processing on the ill word entities so as to obtain ill word pinyin entities and ill word homophonic entities; and

extracting a relationship among the ill word entities, the ill word pinyin entities and the ill word homophonic entities according to pinyin conversion, homophonic conversion and part-of-speech and term frequency, and constructing triples by entity disambiguating so as to obtain the ill-text knowledge graph;

wherein obtaining the first recognition result specifically comprises:

screening the entities of the basic text corpus according to the ill-text knowledge graph so as to obtain a plurality of candidate ill entities by a preset number; and

mapping the basic text corpus and the candidate ill entities into a multidimensional vector using a word2vec model, calculating similarity between the basic text corpus and the candidate ill entities according to a cosine-similarity calculation method, and obtaining the first recognition result according to the similarity;

wherein the corpus recognition model is a k-nearest neighbor model;

wherein a construction of the corpus recognition model comprises:

acquiring the ill information fed back by users and collecting normal corpus information;

performing pinyin conversion and homophonic conversion on the ill information and the normal corpus information word by word to obtain pinyin corpus information and homophonic corpus information;

dividing the ill information, the normal corpus information, phonetic corpus information and the homophonic corpus information as a sample set into a training set and a test set, and mapping the sample set into a spatial vector through the word2vec model; and

performing training on the training set mapped into the spatial vector using a k-nearest neighbor model so as to obtain the corpus recognition model;