Information processing method and device

Abstract

The present invention discloses an information processing device and method, wherein, the method includes encoding an embedded vector to obtain an encoding vector; re-embedding a partial vector of the encoding vector; performing iterative decoding on the encoding vector and sorting and filtering; performing partial vector re-embedding on the representation vector after the iterative decoding and sorting and filtering. Therefore, the embodiment of the present invention uses the combined function of the feature encoding module and the sorting and filtering module, utilizes the return and reuse of feature encoding, and continuously scores and sorts the encoded information, so as to realize high-quality feedback of feature encoding to the input data module.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to an information processing method and device. Background technique [0002] Extracting relational triples (entity-relation-entity) from unstructured text is the key to building large-scale knowledge graphs. Usually, early information extraction regards entity extraction and relationship extraction as a series of tasks. Such a series model is relatively simpler in modeling, but this way of treating entity recognition and relationship extraction as two independent tasks obviously has the following disadvantages , 1. Entity redundancy. Since the extracted entities are paired first, and then the relationship is classified, the redundant information brought by the candidate entity pairs without relationship will increase the error rate and increase the computational complexity. 2. The lack of interaction, ignoring the internal connection and dependency between the two tasks,...

AI Technical Summary

Problems solved by technology

Usually, early information extraction regards entity extraction and relationship extraction as a series of tasks. Such a series model is relatively simpler in modeling, but this way of treating entity recognition and relationship extraction as two independent tasks obviously has the following disadvantages , 1. Entity redundancy, since the extracted entities are paired first, and then the relationship is classified, the redundant information brought by the candidate entity pairs without relationship will increase the error rate and increase the computational complexity

2. Lack of interaction, ignoring the internal connection and dependency between the two tasks, resulting in the result of the relationship extraction task being heavily dependent on the result of entity extraction, leading to the problem of error accumulation

However, in this case, a trade-off needs to be made between the richness of sub-model features and the accuracy of joint decoding: on the one hand, if an accurate joint decoding algorithm is designed, it is often necessary to restrict features, such as modeling with conditional random fields, The global optimal solution can be obtained by using the Viterbi decoding algorithm, but it is often necessary to limit the order of the features, resulting in a reduction in the richness of the sub-model features

On the other hand, if an approximate decoding algorithm is used, such as beam search, features of any order can be extracted in terms of features, but the results obtained by joint decoding are inaccurate

Images