A Representation Method of Event Knowledge Based on Behavior

Abstract

The invention discloses a behavior-based event knowledge representation method, which relates to the technical field of event knowledge representation. The key points of the technical solution are: taking the behavior of driving events as the core, symbolizing the text representation of the event through behavioral algebra and abstract grammar, And build a Behavior Base GCN model with behavioral base layer, action layer and event layer from bottom to top, and convert structured information into vector representation through the model. The method of the present invention guarantees the completeness of the behavior set by studying the homomorphism and isomorphism between behavioral algebras under the deduction of mathematical theory, and uses the graph representation learning framework of the GCN model to construct the representation and representation of events layer by layer from the behavior base. The relationship between events and events has strong generalization and robustness.

Description

technical field [0001] The present invention relates to the technical field of event knowledge representation, and more particularly, to a behavior-based event knowledge representation method. Background technique [0002] An event usually refers to the occurrence of an action or situation involving participants, or a change in the state of an event. Formally, the constituent elements of an event usually include the trigger word or type of the event, the participants of the event, the time or place of the event, and so on. In the real world, events are usually a more structured representation of information. The existing research mainly has the following problems: [0003] (1) There is sparsity, which is not conducive to large-scale representation: [0004] Early research mostly adopts discrete event representation, usually representing events as tuples consisting of event elements. For example, an event is represented as a triplet of a collection of objects, a relations...

AI Technical Summary

Problems solved by technology

However, the biggest problem with discrete representations is sparsity. It is difficult for us to represent large-scale events through such tuples.

[0005] (2) Situational semantic relevance and event interaction are difficult to balance:

The commonly used method is to directly vectorize the words in the text. The disadvantage of this method is that events with low correlation but highly similar texts are expressed similarly, for example: "She throws football on the playground (she throws football on the playground) " and "She throws bomb on the playground (she throws bombs on the playground)" will get a similar vector representation, although the two events are not semantically similar

[0007] (3) Loss of semantic information and poor generalization:

However, related research needs to use predefined domain-specific event patterns to extract real-world events and supplement specific information to ensure the performance of downstream training tasks. Event representations learned in this way often lack good generalization capabilities.

Images