Representing Verbs As Argument Concepts

With the application of artificial intelligence technology more and more in-depth,artificial intelligence technology plays a more important role in human life.Natural language understanding technology is one of the most important techniques for computer to understand human language.The most difficult part is understanding the semantics of human language.Verbs play a central role in understanding the syntactic and syntactic aspects of sentences.The distribution hypothesis indicates that the context information of a word can be used to represent the semantics of the word,such as the words surrounding the word.In recent years,the a lot of use of Word2 Vec technology is the application of the distribution theory,but Word2 Vec only applied the linear context of the word relationship.A verb has its unique role in the sentence,because it contains the subject and object with its dependent relationship,so you can use the verb subject and object to express the verb semantics.Similar models include ReVerb,which uses the ”bag-of-words model” approach,but this model has the following disadvantages: 1)can not consider the relationship between synonyms;2)”bag-of-words model” dimension is very high,so the calculation efficiency is low;The resulting model is human unreadable.In order to compensate for these shortcomings,a natural way is to use these subjects and objects of the abstract concept or type to represent,rather than directly use these words.A similar system includes FrameNet,which expresses the verb's semantics by labeling the verb's subject and object types.However,this system also has some obvious shortcomings: 1)manual labeling workload is huge,so can not expand;2)verb concept is too abstract,for example,the verb ”eat” object concept only contains ”Ingestibles”,so can not express verbs multiple semantics.In view of the shortcomings of the prior art,the purpose of this study is to provide a method and system for expressing verb semantics based on argument concepts.Compared with the existing research,this work has the following beneficial effects: 1)provides a method to express verb semantics based on argument concepts,creatively utilizes an external knowledge base to express verbs' semantics,and provides the user with the parameter of the semantic granularity of the verbs,so that the semantic concept of verbs is moderate;2)The semantic concept of verbs which can be read by human can be obtained by the method of expressing verbs' semantics based on the concepts of arguments,which can also be directly calculated by machine.This paper defines the problem of ”Argument Conceptualization”: Given a set of argument instances(subject or object)of the same kind of a verb,we want to extract k concepts from the external concept-entity knowledge base,which under the ”isA” relationship can cover as much as possible the above parameters set.At the same time,we hope to extract out of the k concepts between which the ”semantic coincidence degree” can be as small as possible.From the problem of ”Argument Conceptualization”,this paper defines the ”Concept Graph” model and transforms the problem into a formalized algorithm problem of ”finding the largest k-cluster of weighted sums in weighted graphs”.And it is proved that the problem is an NP-Complete problem,and then an efficient search algorithm is proposed to solve it.Finally,the paper proves that the concept of the argument concepts deduced from the proposed algorithm has excellent performance through manual review.For the three objective natural language processing task-Verb Argument Identification,Verbs Clustering and Term Similarity Computation.In the evaluation results,we can find that the proposed algorithm gives a comprehensive improvement of the conceptual results of the arguments compared with the contrast method.