Research On Question Answer Intention Recognition Algorithm With Knowledge Graph

Internet information is increasing in large quantities every day.Faced with the massive amount of information,it is very important to accurately identify the user’s intention of the problem and provide users with accurate and useful information.Intent recognition has been widely used in question answering platforms,search engines,information retrieval,online medical diagnosis and treatment and other fields.Intent recognition is mainly divided into intent classification(recognition)and unknown intent recognition(detection).Capsule network inherits the advantages of CNN,improves its limitations,and gradually becomes a commonly used method in intent classification research.However,the existing capsule network models for intent classification have certain limitations in terms of long-distance dependent encoding in text sequences and in selecting and paying attention to important words.In order to further improve the effect of capsule networks for intent recognition,this paper combines attention mechanism,parallel Routing technology and knowledge graph are integrated with capsule network to conduct research on intent recognition.The main research work of this paper is as follows:(1)A non-iterative parallel routing capsule network model NIPR-AT Capsule Net based on attention mechanism is proposed.First,the attention mechanism is combined with the capsule network to calculate the attention value of the utterance,so as to improve the efficiency and accuracy of the subsequent module processing sentences;secondly,a non-iterative,parallelized routing algorithm is introduced to form text core words into text Represents and encodes the dependencies between text sequences;finally,integrates domain knowledge graph information to achieve information enhancement in intent recognition,further improving the accuracy of the model.Comparing with the Text CNN method on the Subj dataset,the experimental results show that the NIPR-AT Capsule Net improves by 1.9%,4.9% and 1.4% in the three indicators of ER(Exact match rate),Recall and F1,respectively.(2)A capsule network model based on the zero-trigger mechanism is proposed to identify unknown intentions.The model consists of three types of capsules:Semantic Caps,Detection Caps and Zero-shot Detection Caps.Among them,Semantic Caps combines capsule network with LSTM to extract semantic features from sentences;Detection Caps aggregates the semantic features extracted by Semantic Caps to form intent labels;Zero-shot Detection Caps effectively identifies emerging intents through a knowledge transfer strategy.In order to obtain the knowledge in the knowledge graph,the ZSL-KG framework is introduced,and the information in the knowledge graph is used for information enhancement and supplementation for intent recognition.Comparing experiments with TFIDF-LR and other methods on public datasets,the experimental results show that the proposed capsule network model is effective.(3)By applying the intent recognition model proposed in this paper to a knowledge-enabled question answering platform,the potential of the question and answer intent recognition algorithm based on knowledge graph proposed in this paper in practical scenarios is demonstrated with practical application effects.