Multi-output text classification model construction method and device, equipment and medium

Abstract

The invention discloses a multi-output text classification model construction method and device, equipment and a medium, and the method comprises the following steps: obtaining original text data in aproperty forum, and carrying out the classification marking; finding out a phrase combination with the maximum word frequency in the classified and marked sentences; gathering all words in the phrasecombination with the maximum word frequency, and converting the gathered words into an One-Hot form; using a text corpus based on a property forum for training a CBOWnegative sampling unsupervised model of Word2Vec (Word2Vec); importing the word set converted into the One-Hot form into a Word2Vec model, and outputting a Word Embedding vector; constructing two CNN neural network models based on the same Word Embedding vector, and sharing the same input word vector; and enabling concate to merge the two CNN neural networks, and respectively use different full connection layers to classify a plurality of different categories. According to the invention, quick response to user demands is realized, the flow processing time is saved, the complaint type is quickly discriminated, junk advertisement content is effectively screened, and the property service quality is improved.

Description

technical field [0001] The present invention relates to the field of text classification, and more specifically relates to a construction method, device, equipment and medium of a multi-output text classification model based on property forum content. Background technique [0002] The following information mainly exists in the content of the property user forum: 1. Category: complaint / advertisement / maintenance / commercial / service / safety; 2. Sentiment: positive / negative / neutral. By correctly classifying content categories, it is helpful to determine and generate work orders, and assign corresponding professionals to solve them. By correctly judging the emotions expressed by the content. [0003] However, at present, the content and information of property user forums are all manually classified, and manual classification has the following defects: 1. Through the owner's own report of the work order, the active screening and selection of the report category often exists and us...

AI Technical Summary

Problems solved by technology

[0003] However, at present, the content and information of property user forums are all manually classified, and manual classification has the following defects: 1. Through the owner's own report of the work order, the active screening and selection of the report category often exists and uses the default option, which is accurate The degree is poor, and manual housekeepers are required to intervene to screen and initiate reports again, which increases the service process and does not provide a better user experience

2. Through the manual judgment and classification of property operations, manual screening of text content cannot play a role in saving labor costs, and manual screening and classification also have certain subjective judgment errors, which brings unnecessary process redundancy

3. There are a large number of advertisements and spam content in the forum, and the amount of reading is large, and manual processing is easy to miss, resulting in real and useful customer needs not being processed

[0004] Based on the above-mentioned defects of manual classification, it can be concluded that in the current property user forums, the massive service feedback from customers is not processed in a timely manner, time-consuming to filter, difficult to classify appeal categories, difficult to filter garbage content, and the degree of automation of the processing and response mechanism not high

Images