A method for entity relations extraction including applying entity markers to a set of sentences included in a data bag to generate a token sequence for a subset of the set of sentences, the token sequence including a beginning position mark and an ending position mark of a corresponding sentence, as well as a front position mark and a rear position mark of at least one entity included in each of the subset of the set of sentences; using the generated token sequences of the set of sentences with a pre-trained language representation model to generate a sentence feature vector for each sentence included in the data bag; aggregating, in a data encoding module, the sentence feature vectors of the set of sentences into a bag encoding vector; and classifying data entity relations of the set of sentences included in the data bag through decoding and inferencing the bag encoding vector.