Research On FAQ Question And Answer System In Over-All Ship Performance Field

In recent years,with the development of information technology,the shipping industry has also gradually started transforming information.Ship design,production,and maintenance have all started to be digitized and automated,and the internal management of enterprises has also increasingly evolved toward the tip.Some new technologies,such as artificial intelligence,deep learning,and knowledge graph,have promoted the information of the shipping sector and the ease of retrieving data.Traditional information retrieval methods often cannot meet users’ needs for domain-specific knowledge queries,making the question and answer systems based on natural language processing technology an essential tool for solving such problems.The main task of an FAQ system is text semantic matching,retrieving similar questions from the knowledge base and returning the corresponding answers,but the current text semantic matching model does not learn the deep semantic information in the users’ questions very well.In addition,the amount of data available for FAQ question and answer pairs in a given domain is small and the cost of manually constructing question and answer pairs dataset is hard,so data augmentation is also a consideration in the FAQ system.Based on the practical application scenarios of the Ship Overall Performance Innovation Research Open Fund,this thesis researches and designs a FAQ system for the field of ship overall performance,which is used to solve the knowledge retrieval needs of engineers for specific domains.The main research elements of this thesis are as follows:(1)A data enhancement method for sentence paraphrase generation that incorporates syntactic information and edit vectors is proposed.Firstly,the template sentences corresponding to the original sentences are retrieved from the corpus using a rule-based approach,followed by the lexical analysis of the template sentences using the spa Cy tool.Then special characters mask the relevant lexical words,such as nouns,verbs,adjectives,and adverbs in the template sentences.Finally,we used the Glove embedding to construct an edit embedding between the original sentence and the reference paraphrase sentence,and the edit embedding was added to the encoding layer of the pre-trained model to enhance the model’s learning of the differences between the original sentence and the reference paraphrase sentence.In this thesis,the effectiveness of the paraphrase generation model is verified on both English and Chinese datasets by means of evaluation metrics and manual evaluation.(2)A mixed feature text semantic matching model based on Sim BERT is proposed.The keyword features in the text are first extracted using a keyword extraction tool,then the intention features are constructed by replacing the keywords in the text with special characters.Finally,the user questions,keyword features and intention features are joined as the model input.In addition,this thesis constructs a corpus of professional dictionaries in the ship domain and FAQ question and answer pairs in the ship general performance domain through crawlers and data cleaning techniques.It validates the model performance on the public dataset and the ship general performance domain dataset.(3)Design and implement a FAQ question and answer system for the general performance domain of ships.A separate front and backend development framework of Vue and Flask.The system deploys deep learning models with the Tensorflow Serving tool.The system provides automatic question and answer and history record queries,management of resources such as FAQ question and answer pairs,dictionaries,and vector index files,online data annotation,model training,and model update.Finally,package the interface of the FAQ system and test the system pressure to ensure the regular operation of the system under high load and other conditions in real applications.