Demand extraction base on conversation

Aim:

.The demand trend detect of client conversation with employees
.Demand extraction and apply for future agents reminder under task handle circumstance or automatically Robot Q&A

word_cut-extract_judge code method

-> Data preprocessing
-> Built StopWord dict and AddWord dict
-> Original Text cut words into list (base on two dicts)
-> Keyword Extraction (base on tfidf or TextRank)
-> Rule dictionary built (like {"Tag":[k1,k2,k3]} currently by manual)
-> Sentences Text import and Judge module apply

Stage 2 been Done

.N_gram method for high probability rule dictionary update
.N_gram to N_follow judge fix match to dynamic match, aiming to provide a wider demand construction

Next Stage and issue

.Low Recall on all terms especially on Covid term
.Futrue Plan: Add more realted terms to fill the gap, About short term trend. considering single words bags to detect. As for long term trend, considering realted words bags for detection.

Future Plan

Automatic Model to detect high probability words match