Research On Key Technologies Of Intelligent Question Answer System In E- Commerce Domain

With the rapid development of Artificial Intelligence and Robotic Techniques, a QA system is capable of providing an accurate answers and intelligent services to the end user and has become a hot research field.It is gradually entering the work environment as a substitute to manual work as commercial services switch from offline to online. With costs soaring in Call centers implementing, intelligent QA system can help save in the region of 80% of the manual work. In view of this, the research about QA system in E-commerce domain has great application requirements and practical value.The key of the intelligent QA systems is understanding the needs of the end user and providing the solutions that meet their business needs. QA systems which are primiarily based on structured data are the mainstreaming trend now. However, Natural language has ambiguity and does not follow the syntax. To convert Chinese oral language into logical query is an important and challenging task to study. Knowledge can help QA system becoming more intelligent. However, in order to build a large scale of knowledge base requires a lot of manual work. The precision of information extraction is low. The knowledge graph exists semantic fuzzy, data missing, uncertainty and conflict. So, how to build clear and high quality large scale knowledge base is a difficult task. Intelligent QA systems with self-learning capabilities would be the trend in the future.Intelligent QA system has a strong domain sensitive. How to enhance semantic understanding, knowledge representation and reasoning ability is the key issue of the QA system. According to the actual requirement and the existing problems, the paper proposed a Chinese semantic parsing method for automatic template generation and related algorithms are improved. For the knowledge base construction, the improved knowledge representation model and the automatic fusion and prediction method are put forward. The intelligent QA system is able to answer the questions of reasoning, statistics etc. Concrete work is as follow:Due to the complexity of domain term, both English and Chinese is merged in the same sentence, which can lead to the part-of-speech tagging error and the introduction of noise. This paper proposed an entity recognition algorithm based on the CRF to recognize the mixed text of English and Chinese noun, which data is trained by the title data of E-commerce websites, the recognized accuracy is as high as 95%. This paper puts forward the integrated Word2vec feature relation extraction algorithm based on CRF, which improves the accuracy of the extract synonyms. They are the most important foundation of knowledge fusion and questions semantic understanding.In order to improve the recall rate of the semantic parsing based on template, the paper puts forward Chinese question template automatic generation of semantic parsing method. As the result of question semantic dependency parsing is too complicate, the paper put forward SDP-Reduce method, parsing accuracy increased 40%. By using Word2vec semantic features to improve the entity link accuracy for diversified expression. The precision of the whole CQPT algorithm outperforms the baseline algorithm almost by nearly 80%.The existing knowledge graphs are semantic fuzzy, complex and difficult to extend. This paper puts forward the framework of the inductive-deductive framework, which combines the advantage of ontology and knowledge graph. The CyberSchema Model is proposed based on the core of the ontology knowledge base which is composed of concept, relation and rule triples. The CyberSchema is simple, easy to expand and domain independent, all of which can improve the efficiency and certainty of semantic expression.Because the data is from multiple heterogeneous websites, which is uncertain and can be conflicting, the knowledge fusion algorithm based on Graph model is proposed to establish more unified, complete, accurate attributes and values. The attributes matching algorithm based on semantic and the reliability optimal decision algorithm are brought forward to generate optimal commodity attributes and each entity attribute value automatically.A knowledge base completion algorithm based on tensor and word embedding is proposed to meet Cyber knowledge base requirements. Knowledge base contains a large amount of implicit relationships, this need to be automatic reasoning which can help complete the knowledge base and can answer analogy, reasoning and other complex problems.Finally, based on these algorithms, Chinese intelligence QA system on E-commerce domain is proposed, which composes of four modules: semantic understanding, knowledge extraction, knowledge integration and knowledge completion. Not only does this raise the accuracy of terms recognition and semantic parsing, but also effectively improves the certainty and consistency of fine-grained knowledge expression.The proposed QA system is able to answer the questions about statistics, analogy and reasoning etc. with its effectiveness validated through experiments.