Research On Recommendation Of Famous TCM Cases Based On Similarity

The medical records of famous traditional Chinese medicine practitioners are the accumulation of their experience,carrying the academic thinking and clinical diagnosis and treatment experience of famous traditional Chinese medicine practitioners,and can provide reference for Chinese medicine learners.Finding similar medical cases from a large number of traditional Chinese medicine cases can provide reference for clinical doctors,improve their decision-making level,and effectively promote the inheritance and application of traditional Chinese medicine knowledge.The article is based on twin networks and pre trained models,and further optimizes the structural details of the model based on the characteristics of experimental data and actual matching results.Multiple models are fused to distinguish the similarity of the main complaint text of famous traditional Chinese medicine cases,improving the accuracy of similarity discrimination.The research content and innovative points of this article are summarized as follows:(1)In response to the problems of numerous professional terms,relatively complex language structures,and complex contextual semantic relationships in the texts of famous traditional Chinese medicine medical cases,a matching model for traditional Chinese medicine medical case main complaints that integrates twin networks and deep learning is proposed and applied in recommendation scenarios.This method introduces deep learning models and pre trained models into parallel structures to enhance the model’s efficiency,and can effectively represent medical record text features.The use of multiple attention heads in two representation models to encode contextual information more effectively not only eliminates noise in different text languages,but also improves the overall generalization performance of the model;Utilizing the TCNN structure for secondary feature extraction to enrich feature representation,and fully utilizing the result vector group of the randomly adjusted mask model for multi-channel convolution,further enriching the relationship feature representation between text tokens and improving feature utilization;Optimizing training tasks to fit the characteristics of traditional Chinese medicine medical record texts and adopting a more suitable full word mask training method can help to understand medical record texts at a larger granularity,enhance the effectiveness of MLM training tasks,and thereby improve judgment accuracy.(2)In view of the complexity of the deep semantic relationship between coded entities in TCM medical records,the conventional text semantic matching methods do not fully extract the various deep semantic relationships between entities,which makes the model have greater limitations in the recommended scenario of TCM medical record matching,and cannot fully explore the Semantic information between texts to complete the matching task.This paper proposes a matching recommendation model based on semantic enhancement for the chief complaint text of traditional Chinese medicine cases,constructs a text deep Semantic information learning network based on attention mechanism,fully absorbs the deep semantic relationship information between entities that may exist in sentences,and then uses vector reconstruction to select and enhance the semantics,and finally gets the matching recommendation results through the matching output layer.Using semantic enhanced multi granularity encoding to encode text,adopting three granularity levels of learning: word,structure,and semantics,can more deeply align with the language characteristics of traditional Chinese medicine medical record texts,which is conducive to more accurate allocation of bidirectional attention in the text and obtaining accurate semantic enhanced features;Two representation networks were used for information representation,using a universal representation module and a task specific representation module to extract representations of the text,taking into account both surface general language features and deep level specific task features.On the basis of how to allocate attention,the problem of how much attention should be allocated was solved to better match the actual task scenarios in the field of traditional Chinese medicine medical records;Self encoder is used for vector reconstruction to achieve feature selection: the combination of ERNIE and other pre training models and deep network models is used to build models,and self encoder is used to achieve feature selection and dimension reduction of sentence vectors,so that the matching recommendation results are more accurate and effective.(3)In order to better inherit and apply this study at the grassroots level,based on the practical needs of TCM grassroots clinical practice,and fully combining the previous two parts of research,a TCM medical case matching recommendation system with model result fusion was designed and formed.Weighted fusion and voting fusion methods were adopted to better achieve the medical case matching recommendation task.