BERT-TextCNN: Multi-label Classification & Sentiment Analysis

A dual-task NLP system combining BERT embeddings with TextCNN for:

• Multi-label Text Classification - Assign multiple category labels to each input
• Sentiment Analysis - Classify sentiment as Positive / Neutral / Negative

This project integrates and improves upon my previous works:

• TextCNN_CLS - Single-label text classification
• LCF-ATEPC_ABSA - Aspect-Based Sentiment Analysis (ABSA)

Dataset

The dataset consists of e-commerce product Q&A data from Taobao and JD.com platforms.

Classification Labels (12 categories):

ID	Label	Description
0	功效	Efficacy
1	适用人群	Target Users
2	使用方法	Usage Method
3	其他	Other
4	属性	Attributes
5	使用感受	Usage Experience
6	不良反应	Side Effects
7	竞品对比	Competitor Comparison
8	包装	Packaging
9	价格	Price
10	渠道	Channel
11	物流	Logistics

Sentiment Labels (3 classes):

ID	Label	Description
0	中评	Neutral
1	好评	Positive
2	差评	Negative

Features

Module	Task	Output
TextCNN_Single	Single-label classification	12 categories
TextCNN_Multi	Multi-label classification	Multiple labels per input
TextCNN_Sentiment	Sentiment analysis	Positive / Neutral / Negative
online	Production inference API	Batch prediction support

Quick Start

# Environment setup
conda activate nlp

# Training
cd TextCNN_Single && python train.py
cd TextCNN_Multi && python train.py
cd TextCNN_Sentiment && python train.py

# Inference
cd online && python demo.py

API Usage

# Classification (Multi-label)
from models.classification.functions import predict
label_ids, label_texts = predict(['保湿效果怎么样？', '小孩子可以用吗？'])

# Sentiment Analysis
from models.sentiment.functions import predict
label_ids, label_texts = predict([['是正品吗？', '效果不错']])

Project Structure

├── TextCNN_Single/      # Single-label classification training
├── TextCNN_Multi/       # Multi-label classification training
├── TextCNN_Sentiment/   # Sentiment analysis training
└── online/              # Production inference module
    └── models/
        ├── bert-base-chinese/   # Pre-trained BERT
        ├── classification/      # Classification inference
        └── sentiment/           # Sentiment inference

Tech Stack

• Model: BERT (bert-base-chinese) + TextCNN
• Framework: PyTorch, HuggingFace Transformers
• Environment: Python 3.12, CUDA supported

---

日本語版 README