Multi-lingual health information intelligent customization method and system based on user portrait

Abstract

The application provides a multi-lingual health information intelligent customization method and system based on user portrait, relates to the technical field of health information processing, and comprises the following steps: collecting user feature data to construct a user portrait vector, acquiring health information source data to generate a layered knowledge expression matrix, performing language suitability detection to construct a cross-lingual semantic mapping atlas, and generating a customized health information sequence according to a health demand change trajectory. The method can realize personalized customization and multi-lingual accurate expression of health information, and improves information acquisition efficiency and understanding accuracy.

Claims

1. A method for intelligent customization of multilingual health information based on user profiles, characterized in that: include: Collect user feature data and perform normalization processing to obtain the user basic feature set; The user's basic feature set is mapped to the semantic space to obtain the user's semantic representation. A feature association matrix is ​​constructed in the semantic space to divide the knowledge feature subspace and the language feature subspace. The knowledge feature subspace and the language feature subspace are weighted and fused to obtain the user profile vector. Acquire health information source data and construct a health information entity association graph. Determine the knowledge unit decomposition granularity based on user profile vectors, reconstruct the knowledge units into a multi-level knowledge interpretation sequence, establish hierarchical association relationships, and generate a hierarchical knowledge expression matrix. The hierarchical knowledge representation matrix undergoes language adaptability testing to identify semantic nodes to be transformed and label their transformation sensitivity. Equivalent representations in the target language are generated, semantic verification relationships are established, and a cross-language semantic mapping graph is constructed, including: Extract semantic feature vectors of expression units in the hierarchical knowledge expression matrix, calculate the distance between the semantic feature vectors of the source language and the target language, determine the semantic node to be converted when the distance exceeds the preset distance threshold, calculate the feature distribution variance of the semantic node to be converted in different language spaces, and determine the conversion sensitivity. Based on transformation sensitivity, the hierarchical relationship of the semantic nodes to be transformed in the hierarchical knowledge expression matrix is ​​extracted. Precursor expression units and successor expression units are extracted according to the hierarchical relationship, and a context semantic vector is constructed. The context semantic vector is used as a constraint to retrieve candidate expressions in the target language space. The semantic similarity and context matching degree of the candidate expressions are calculated. The candidate expressions that simultaneously meet the preset similarity threshold and the preset matching degree threshold are selected to determine the equivalent expression in the target language. Perform a back-translation on the equivalent expression in the target language to obtain the back-translated expression. Calculate the semantic deviation between the back-translated expression and the semantic node to be converted, including: Extract the lexical sequence of the equivalent expression in the target language, find the corresponding source language lexical in the target language lexical space for each lexical in the lexical sequence, reorganize the found source language lexical according to the grammatical structure of the equivalent expression in the target language to obtain the initial back-translation expression, extract the contextual dependencies of the initial back-translation expression, adjust the word order and grammatical correction of the initial back-translation expression according to the contextual dependencies, and determine the back-translation expression. Extract the first semantic feature vector of the back-translated expression, extract the second semantic feature vector of the semantic node to be converted, calculate the vector angle and vector magnitude difference between the first and second semantic feature vectors in the semantic space, and calculate the semantic deviation value in combination with the preset corresponding weight coefficients. When the semantic deviation value is less than the preset consistency threshold, establish a semantic verification relationship. By using the semantic nodes to be converted and the equivalent expressions in the target language as graph nodes, and the semantic verification relations as graph edges, a cross-language semantic mapping graph is constructed. Based on the user's basic feature set, the trajectory of health demand changes is calculated. Then, the cross-language semantic mapping graph is sorted and prioritized according to the trajectory of health demand changes to obtain a customized health information sequence. Customized health information sequences are pushed to user terminals.

2. The method according to claim 1, characterized in that, Constructing a feature association matrix in the semantic space to divide the knowledge feature subspace and the language feature subspace includes: A bidirectional propagation analysis is performed on the features in the semantic space. The forward propagation calculates the semantic relevance between feature nodes to obtain the first feature association matrix. The back propagation iteratively optimizes the association weights based on the activation degree of feature nodes to generate the second feature association matrix. Based on the second feature association matrix, a feature propagation network is constructed, and the first-order neighbor features and second-order neighbor features of each feature node are fused through a multilayer perceptron to obtain the third feature association matrix. Based on the third feature correlation matrix, calculate the mutual information and conditional entropy between feature nodes, construct a feature propagation graph, perform node clustering on the feature propagation graph, and obtain the initial feature subspace partitioning result; The initial feature subspace partitioning result is subjected to feature vector decomposition. Based on the symbol structure of the feature vectors, the feature nodes corresponding to the positive feature vectors are partitioned into the knowledge feature subspace, and the feature nodes corresponding to the negative feature vectors are partitioned into the language feature subspace.

3. The method according to claim 1, characterized in that, Acquire health information source data and construct a health information entity association graph. Determine the granularity of knowledge unit decomposition based on user profile vectors, reconstruct knowledge units into multi-level knowledge interpretation sequences, establish hierarchical association relationships, and generate a hierarchical knowledge representation matrix, including: A semantic feature extraction network is used to identify health information entity words and relation words from health information source data, and a health information entity association graph is constructed. Based on the health information entity association graph, knowledge points and attribute information are extracted, and the semantic dependency degree between health information entities is calculated to generate knowledge dependency links. Based on the knowledge dependency links, a decoupling matrix is ​​constructed to deconstruct health knowledge content into a set of knowledge units; Based on the user profile vector, the semantic similarity with each knowledge unit in the knowledge unit set is calculated to obtain the knowledge unit comprehension level value, and the knowledge unit decomposition granularity value is determined according to the knowledge unit comprehension level value. Semantic reconstruction is performed on knowledge units according to the knowledge unit decomposition granularity value to generate a multi-level knowledge interpretation sequence, construct a knowledge progression path graph, calculate the knowledge transfer probability between hierarchical nodes in the knowledge progression path graph, construct a knowledge transfer network for hierarchical nodes that meet the transfer threshold, and generate a hierarchical knowledge expression matrix based on the connection structure of the knowledge transfer network.

4. The method according to claim 3, characterized in that, The semantic reconstruction includes: The knowledge unit is semantically decomposed, and health information concept words, modifiers and limiting words are extracted to construct a semantic component library. The semantic complexity of word combinations is calculated based on the semantic component library. The word combinations are classified according to the semantic complexity value to generate a classified word sequence. The semantic difference value between adjacent levels in the classified word sequence is calculated. It is determined whether the semantic difference value exceeds a preset cognitive threshold. The position that exceeds the cognitive threshold is marked as the semantic breakpoint position. Insert supplementary explanations at semantic breakpoints, replace professional health information terms with general expressions, and generate semantic transition content. Semantic transition content is inserted into hierarchical word sequences according to semantic breakpoints to construct a semantic link. Semantic association values ​​between adjacent levels in the semantic link are calculated. Semantic content whose semantic association values ​​meet the preset continuity requirements is selected and organized to generate a multi-level knowledge semantic sequence.

5. The method according to claim 1, characterized in that, Based on the trajectory of changing health needs, the cross-lingual semantic mapping map is sequentially arranged and prioritized to obtain a customized health information sequence, including: The trajectory of changes in health needs is divided into time-series demand segments according to time nodes. Semantic nodes corresponding to the time-series demand segments are retrieved from the cross-language semantic mapping graph to generate a semantic node sequence. The transition relationship between adjacent time-series demand segments is analyzed, intermediate semantic nodes are supplemented, and an extended semantic sequence is constructed. Calculate the transmission strength between semantic nodes in the extended semantic sequence, generate node transmission links, and arrange the node transmission links in time sequence according to the occurrence time of the time sequence requirement segments to obtain ordered transmission links; The requirement satisfaction and information integrity of semantic nodes in the ordered transmission link are calculated, node priority values ​​are generated, the ordered transmission link is classified according to the node priority values, a hierarchical transmission network is constructed, the temporally corresponding semantic nodes are extracted from the hierarchical transmission network, and customized health information sequences are generated.

6. A multilingual intelligent health information customization system based on user profiles, used to implement the method of any one of claims 1-5, characterized in that, include: The first unit is used to collect user feature data and perform normalization processing to obtain the user basic feature set; The second unit is used to map the user's basic feature set to the semantic space to obtain the user's semantic representation. In the semantic space, a feature association matrix is ​​constructed to divide the knowledge feature subspace and the language feature subspace. The knowledge feature subspace and the language feature subspace are weighted and fused to obtain the user profile vector. The third unit is used to acquire health information source data and construct a health information entity association graph. Based on the user profile vector, the knowledge unit decomposition granularity is determined, the knowledge unit is reconstructed into a multi-level knowledge interpretation sequence, and a hierarchical association relationship is established to generate a hierarchical knowledge expression matrix. The fourth unit is used to perform language adaptability detection on the hierarchical knowledge representation matrix, identify semantic nodes to be converted and mark conversion sensitivity, generate equivalent expressions in the target language, establish semantic verification relationships, and construct a cross-language semantic mapping graph. The fifth unit is used to calculate the trajectory of changes in health needs based on the user's basic feature set, and to perform temporal arrangement and priority assignment on the cross-language semantic mapping map according to the trajectory of changes in health needs, so as to obtain a customized health information sequence. The sixth unit is used to push customized health information sequences to the user's terminal.

7. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 5.

Images