Recommendation Method For Elderly Care Service Based On Knowledge Graph And Sentence Embedding

As a result of the growing aged population,the elderly care service industry is booming.The accurate recommendation method for elderly care service is the key to provide high quality service.The elderly has the characteristics of fixed habits and vague expression of demand.However,the collaborative filtering recommendation method,which is commonly used,doesn't optimize such fields.Therefore,this thesis analyzes the characteristics of the elderly care service,builds the domain knowledge base,designs and develops the recommendation method for the two typical problems.The result proves the correctness and effectiveness of the method.Firstly,this thesis analyzed the characteristics of elderly care service in China,investigated the complex needs in this multi-field services,and summarized the core problem that need to be solved.Due to lack of a well-formed information base in this field,it is impossible to integrate a large number of heterogeneous information.Therefore,this thesis used knowledge graph to integrate heterogeneous information,and construct a large-scale and high-quality Chinese Knowledge Graph for Elderly Care Service(CKGECS),which can help related work.Aiming at the characteristics of the elderly with fixed habits,this paper studied a recommendation method based on knowledge graph for implicit preference with time series.The method firstly selected the deep sequential model to learn the implicit preference from the historical interaction records.By comparing several mainstream recommendation methods,it proved that the optimal recommendation for solving such problem is the Gated Recurrent Unit(GRU)model.However,this method didn't use all the information from the service and the interaction record.Therefore,the improved Embedding Projection for Knowledge Graph Completion(ProjE)was used for learning suitable service vector.After the coding layer wass replaced,the effect of the GRU model was improved,with better service recommendation result.Aiming at the problem of vague and incomplete elderly care service demand,this paper studied a vague requirement service recommendation method based on sentence embedding.The method chose to use sentence embedding to connect the requirement with the service description.After preprocessing,the Global Vectors for Word Representation(GloVe)method was used to obtain the word vector.By improving the Weighted-Removal sentence embedding method,the sentence vector was better measured.The experimental results also proved that it was better than the recommendation method with direct retrieval according to the requirements.Finally,the recommendation tool for elderly care service was designed and implemented,and the proposed methods were verified.The recommendation results of the two methods were combined to meet the user's requirements while mining user preferences,and the recommendation effect was further improved.The application in the real scenario proved the effectiveness and practicability of the method.