BIG-DATA-BASED ZERO ANAPHORA RESOLUTION METHOD AND APPARATUS, AND DEVICE AND MEDIUM

摘要

A big-data-based zero anaphora resolution method. The method comprises: acquiring a sentence to be resolved and preceding text information thereof, and performing vectorization processing on the sentence to be resolved and the preceding text information thereof so as to obtain a context vector representation of each word in the sentence to be resolved and a context vector representation of each word in the preceding text information (S101); inputting the context vector representation of each word in the sentence to be resolved and that in the preceding text information into a bidirectional long short-term memory network to enhance the context expression and position information of each word, so as to obtain an enhanced context vector representation of each word (S102); traversing the enhanced context vector representation of each word, and according to a parameter vector in a bert model, predicting an anaphora item head word probability and an anaphora item tail word probability of each word (S103); traversing each word, constructing continuous text segments, and according to the anaphora item head word probability and the anaphora item tail word probability of each word, calculating anaphora item probabilities of the continuous text segments (S104); and selecting the continuous text segment with the maximum anaphora item probability from among the continuous text segments as an anaphora item of the sentence to be resolved (S105). The method solves the problems of existing zero anaphora resolution techniques being excessively dependent on an anaphora item candidate set, and a resolution result being low in accuracy and being unstable.