Vertical Domain Entity Disambiguation Method Fused with Topic Model and Convolutional Neural Network

Abstract

A vertical domain entity disambiguation method integrating topic models and convolutional neural networks, including: 1. Building a domain knowledge base; 2. Training the preprocessed data set for word vector model, and building a corresponding dictionary; 3. Extracting The name of the entity to be disambiguated, and the candidate entity set corresponding to the entity is determined from the domain knowledge base, and the entity context information is represented by word vectorization; 4. Use the manual annotated training set corpus to construct the key representing the entity subject by using the thesaurus The word dictionary is used as input to train and save the topic model; 5. Use the manually annotated data set as training set and validation set for training, optimize the model parameters and save the CNN model; 6. According to the similarity of the topic features obtained in steps 4 and 5 The degree Sim1 and the semantic feature similarity Sim2 use the weight normalization operation to fuse the two feature similarities in an optimal way; 7. The candidate entity corresponding to the fusion similarity with the largest value is the final disambiguation entity.

Description

technical field [0001] The invention belongs to the technical field of computer data processing, and in particular relates to a vertical domain entity disambiguation method. Background technique [0002] In the Internet age, information is exploding. In the face of massive amounts of information, cutting-edge AI technology can associate text with massive physical information, improve user reading fluency, and achieve precise user experience improvements. Intelligent information processing not only provides intelligent services for specific fields, but also provides more room for innovation. [0003] Entity disambiguation is the core of natural language processing. Its essence is that a word in a sentence may have multiple meanings, and the exact meaning it expresses needs to be determined through context and knowledge base related knowledge. The full name of a company in a specific field is deterministic, but in texts such as information, research reports, and Q&As, the co...

AI Technical Summary

Problems solved by technology

First of all, mainstream methods use entity surface features, popularity features, etc. These features only calculate the similarity between referring items and candidate items from the global context information of the document, ignoring the topical information that is locally obvious in the text, and grasping the topic of referring items There must be a large error; secondly, the bottom layer of the mainstream disambiguation model is mostly based on the bag of words model, which cannot consider the problem of grammar and word order, and it is difficult to reflect the influence and constraints of contextual semantics on entities. As a result, these models cannot make full use of the context to effectively extract semantics feature

Images