Data-mining on 51Job website
This project is maintained by wangjksjtu
This repository is established to explore the data on 51Job website, where a number of companies poster their wanted positions, and at the same time employees could share their own profiles to boost their career development. Overall, the work in this repo could be summarized in following aspects:
The documents of our work are available here: [report], [notebook].
We use scrapy to crawl raw data from 51Job website. See the directory /job51spider
for codes. XPath is used to parse the html and extract data information.
After entering the directory, input the command in cmd.exe to run the spider.
scrapy crawl 51job
We use python libraries pandas (using class dataframe) and re to preprocess the raw data. See /preprocess/preprocess.py for code. You can find the preprocessed data in /data, where middleData.csv is the preprocessed data suitable for drawing pics, while quantityData.csv quantifies all data and fits further data analysis.
See directory /pics
.
We analyzed feature coorelation and feature distribution respectively. We found two some main features which affect salary level: education level requirements, work experience requirements and area location.
Model | R2 value | Mean Error ¥/year | time / s |
---|---|---|---|
Ada-Boost | 0.2350 | 37483.9 | 0.79 |
Grad-Boost | 0.3237 | 34031.4 | 3.13 |
SVR (RBF) | 0.0092 | 43301.9 | 350.08 |
Bayesian Ridge | 0.2667 | 34031.4 | 0.05 |
Elastic Net | 0.0426 | 44784.4 | 0.03 |
MLPs | 0.2682 | 36207.3 | 19.29 |
Model | Accuracy / % | time / s | Model | Accuracy / % | time / s |
---|---|---|---|---|---|
LP | 7.79% | 3.89 | MLP | 20.91% | 20.05 |
GNB | 7.32% | 0.23 | SVM | 29.31% | 1032.75 |
KNN | 25.19% | 2.60 | XGBoost | 27.53% | 303.21 |
RF | 28.44% | 1.80 |
The accuracy & time plot of the above models: