/Event-Classification

use regular expression, SVM, TF-IDF, x2test; classify Chinese news into three events: "financing event"(融资事件), "products release"(产品发布), "other"(其他)

Primary LanguagePythonMIT LicenseMIT

Event-Classification

Dataset

  • total 36497 chinese news, which is collected from the Internet
  • the news in the dataset may be classified into wrong class

Purpose

  • build a classifier, which classifies Chinese news into three events: "financing event"(融资事件), "products release"(产品发布), "other"(其他)
  • use the classifier to classify the news we collect on the Internet, which is stored on mongoDB
  • provide service to a bussiness intelligence product

Method

regular expression

  • use regular expression to analyze the title of the news. For example: if "融资" appear in the title, the news is likely to be "融资事件"
  • clearn data: if the title of "其他" fit the regular expression of other two classes, change the label according to regular expression

support vector classification

  • use jieba to tokenize cleaned data
  • use TF-IDF and x2test to select keywords
  • set the keywords frequent vector as word vector
  • set word vector as X, class of the news as Y, use SVC to implement classification and build classifier

combine SVC & rules

  • use regular expression to clean training data
  • if the title of news fits regular expression, apply it;
  • if the title of news doesn't fit regular expression, use SVC classifier we built

Documents

clean_data

  • loadNews.py: store info of news
  • strongRules.py: use regular expression to analzye title; clean data

get_classifier

  • step1_get_word_vector.py
  • step2_seprerate_train_test.py
  • step3_classification.py

classification

  • classify.py: classify the news in the MongoDB
  • Event_Classifier.pkl: the SVC classifier
  • News_Classifer.pkl: the news classifer (classify news into 17 classes)(method is quite similiar as event classification, as the news label is quite near, we implement mutilabel classification.)

Products


Date: 5/4/2017
Author: Aining Wang