Interactive Topic Modeling Method and Device Considering Expert Knowledge

Abstract

The invention provides an interactive theme modelling method and device in consideration of expert knowledge. The method comprises the following steps: acquiring initial document collection; performing theme extraction on the initial document collection by adopting a potential Dirichlet distribution model LDA to obtain K first theme-word distribution; sorting the words under each theme-word distribution to acquire front L words with large probability value under each theme-word distribution; scoring each word in L words based on an expert database to obtain to-be-distributed probability of each theme; adjusting the probabilities of all words under each first theme-word distribution according to the to-be-distributed probabiligy to obtain the second theme-word distribution of each theme; acquiring a third-theme-word distribution of the initial document collection by using the interactive theme model with the constraint based on the second theme-word distribution. Through the method provided by the invention, the word is scored by the expert, and the accuracy of the third theme-word distribution can be improved.

Description

technical field [0001] The invention relates to the technical field of text data processing, in particular to an interactive topic modeling method and device considering expert knowledge. Background technique [0002] With the development of online social media and e-commerce websites, platforms based on user-generated content (UGC) have become an indispensable part of people's daily life. The development of UGC-based platforms has brought huge business opportunities to enterprises, which can help enterprises discover needs, develop products, and manage customers. [0003] In the process of processing the above-mentioned content in the related art, the topic in the above-mentioned content is usually discovered by means of topic modeling. However, since each topic has different contexts, the same or similar topics or words have different semantics, which makes the semantic information of the final content topic ambiguous. Contents of the invention [0004] Aiming at the d...

Images