Research On Selection And Generation Of Conversation Response Based On Knowledge

Dialogue systems have gained popularity over the past few years,particularly in sectors like customer service,healthcare,and education.Nonetheless,despite their widespread use,these dialog systems still struggle to handle complex conversations and provide appropriate,individualized responses.A conversation system can be made better using knowledge-based technology,and it can produce more accurate and helpful answers if its reasoning process is informed by knowledge of particular subjects or themes.Current approaches could be too dependent on predetermined information sources to be universally applicable or difficult to manage innovative or unexpected scenarios.This article will examine approaches that use contextual knowledge and information to increase the relevance and potency of response generation.The following is the article’s main points of the invention and work content:(1)Knowledge-guided multi-turn conversation response selection model.Focusing on the issues of incomplete filtering of information from injection models,incomplete exploration of prospective semantic data,and incomplete analysis of temporal linkages of existing content.Based on deep multi-matching networks,a suggested multi-turn conversation response model(DMMN)is put forth.The model leverages context and knowledge as queries for possible answers,offers a pre-matching layer after encoding the three,and employs a one-way cross-attention method to filter out knowledge-aware context and context-aware knowledge,respectively,to discover crucial information in both.The step of feature aggregation is complete after dealing with the candidate replies and the first two items.While attention mechanisms with gating are used to mine semantic information between response-based knowledge to improve matching feature information,additional Bi LSTM networks are utilized to capture temporal information between response-based contextual conversation information.Eventually,the representation features mentioned above are combined.The original and revised persona-chat dataset performance evaluation findings demonstrate that the model’s recall rate is further enhanced and the retrieval outcomes are better when compared to previous approaches.(2)Knowledge-guided dialog response generation model.An end-to-end response generation model based on customized knowledge and commonsense knowledge is suggested as a solution to the issue that single-source knowledge does not match current conversation generation needs.It may choose the right common sense and personalized knowledge from the available possibilities to produce responses based on common sense and personalized information.In order to encourage dependencies between personalized knowledge,common sense knowledge,and conversation context,the model incorporates a new interactive matching layer in the encoder-decoder architecture.On the Focus dataset,the model outperforms all baseline models in terms of subtasks and response generation for personalized knowledge selection and commonsense knowledge selection.