Research And Implementation On The Acquisition Of Event Connotation Based On Event Attributes

In the field of artificial intelligence,it has always been recognized that the acquisition of commonsense knowledge is the core problem of artificial intelligence.Commonsense refers to the common understanding that exists between people in daily life.People's ideas change all the time,and commonsense develops constantly.Commonsense knowledge is widely used in natural language processing,computer vision and other fields,but it is implicit,large-scale and domain free,and the understanding mechanism of commonsense knowledge is not clear,so the acquisition of commonsensen knowledge has become a bottleneck problem that limits the development of artificial intelligence.At present,according to the degree of automation,commonsense acquisition methods can be divided into three categories: manual acquisition,semi-automatic acquisition and automatic acquisition.Manual acquisition can obtain the implicit commonsense knowledge,but it costs a lot of manpower and material resources,and relying only on manual acquisition of commonsense will lead to low efficiency and difficult to ensure the completeness and reliability of commonsense;semi-automatic acquisition combines manual acquisition and automatic acquisition,but the premise is to have the support of seed knowledge base,seed knowledge base also needs to spend manpower to build.Automatic acquisition can greatly improve the efficiency of commonsense acquisition,but it is difficult to obtain implicit commonsense.In short,there is no universally accepted good method of acquiring commonsense.Event is the knowledge unit of human understanding of the world,and also the processing unit of commonsense knowledge acquisition.There is a lot of information in the event,and behind that information is a lot of commonsense.In recent years,many scholars,when conducting knowledge processing with events as a unit,lack of systematic research on the internal characteristics of events,the relationship between events and the components of events,and the connotation description and acquisition of the essence of events are not perfect enough.This paper focuses on the study of self-motion events,an important kind of knowledge in event knowledge.Self-motion event refers to a kind of event that "moves the physical position of the self-motion subject in motion",which is one of the most common events in daily life.Although the research on self-motion events has made great progress compared with the past,the existing research on self-motion events is mostly simple description and explanation.At present,some large knowledge bases have not systematically studied the self-motion events.The main research contents of this paper include the following three points:(1)Extension of FSTD.In the actual research,we found that FSTD(Framework of Semantic Taxonomy and Description,FSTD,Semantic classification and Description Framework)has been unable to better meet the requirements of semantic analysis and commonsense acquisition.For this reason,this paper expands the self-motion event framework.Based on the semantic roles strictly defined in the How Net knowledge net self-motion event semantic framework and the frame elements defined by the Self?motion framework in Frame Net,this paper summarizes the more perfect self-motion event attributes.Then,K-means clustering algorithm is used to cluster the preprocessed self-motion event definition to get the attribute value,so as to construct the self-motion event attribute library.Finally,the attribute and attribute value of the self-motion event are added to the corresponding self-motion event framework in FSTD in the form of "attribute: attribute value".(2)An automatic classification method of self-motion events based on the attributes of selfmotion events is proposed.Event classification is the basis of event research.Different events have different characteristics and semantically similar events not only have similar characteristics but also contain similar commonsense.Therefore,it is necessary to classify all self-motion events.Based on the occurrence frequency of self-motion event attribute in all self-motion events,this paper classifies self-motion events automatically,and puts forward inheritance rules among self-motion events,and divides events into child events and parent events.The child event can inherit not only the parent event's grammar,but also the parent event's commonsense.(3)Define the semantic constraint rules reflecting the connotation of self-motion events,and verify the correctness of the rules.The semantics of events do not exist alone,but are interdependent with the semantics of other events.The semantic constraints of events reflect the relationships or associations that must exist between events.There are some semantic constraints among the attributes of self-motion events,and there are also some semantic constraints of event relation classes among different self-motion events.Therefore,this paper studies the semantic constraints of the event to obtain the relevant commonsense of the event,summarizes the different types of semantic constraints,and analyzes the possible errors and gives the corresponding detection methods.