The invention belongs to the field of comment information intelligent
processing, and relates to a text feature enhancement technology, an information topic extraction technology, a short text aggregation technology and a text
sentiment analysis technology, in particular to a comment information intelligent classification method fusing long text topic keywords. Comprising the following steps: S1, extracting a topic word vector and a comment word vector of a comment information text by utilizing a BERT model, and enhancing a comment information topic feature by utilizing a topic attention layer; s2, performing dimension reduction operation on the comment vectors by adopting a UMAP
algorithm, and clustering comment texts by adopting an HDBSCAN
algorithm; s3, constructing a comment theme core word set, and aggregating the comment short texts to obtain a comment long document; s4, performing comment topic extraction by adopting a ProdLDA neural
topic model; s5, converting the comment text into two different types of
word embedding through a GloVe model and a pre-trained BERT model; s6, performing local
feature extraction and context
information capture on the text by using a multilayer neural
network model; and S7, realizing
binary classification detection on the comment text by utilizing the sentiment classification model. According to the method, an optimization model
algorithm is provided under a specific scene task, and high-consistency comment information theme mining and high-precision comment information intelligent classification are realized.