# Medical search Query relevance judgment

## Question description

Correlation between queries (i.e., search terms) evaluates how well two queries match the topics expressed by them, that is, whether and to what extent Query-A and Query-B are escaped. The topic of Query refers to the focus of query, and determining the correlation between two query terms is an important task, which is often used in the search quality optimization scenario of long-tail query. This task data set is generated under such background.

<div align=center>

![examples](./pic/1.png)

</div>

## Dataset introduction

[Download](https://tianchi.aliyun.com/competition/entrance/532001/information)

The correlation between Query and Title is divided into three levels (0-2). 0 is the worst correlation, and 2 is the best correlation.

2 points: indicates that A and B are equivalent, the expression is completely consistent

1 score: B is the semantic subset of A, and B refers to A scope less than A

0 score: B is the semantic parent set of A, B refers to A range greater than A; Or A has nothing to do with B semantics

## Structure

```

·

├── data

│   ├──  example_pred.json

│   ├── KUAKE-QQR_dev.json

│   ├── KUAKE-QQR_test.json

│   └── KUAKE-QQR_train.json

├── tencent-ailab-embedding-zh-d100-v0.2.0-s

│   ├── tencent-ailab-embedding-zh-d100-v0.2.0-s.txt

│   └── readme.txt

├── chinese-bert-wwm-ext

│   ├── added_tokens.json

│   ├── config.json

│   ├── pytorch_model.bin

│   ├── special_tokens_map.json

│   ├── tokenizer_config.json

│   ├── tokenizer.json

│   └── vocab.txt

├── pic

│   └── 1.png

├── scripts

│   ├── inference.sh

│   ├── eval.sh

│   └── train.sh

├── train.py

├── eval.py

├── models.py

├── inference.py   

└── README.md

```

## Environment

```shell

pip install gensim

pip install numpy

pip install tqdm

conda install torch

pip install transformer

```

## Prepare

Download corpus from Tencent AI Lab

```shell

wget https://ai.tencent.com/ailab/nlp/zh/data/tencent-ailab-embedding-zh-d100-v0.2.0-s.tar.gz # v0.2.0 100 demention-Small

```

Decompress the corpus

```shell

tar -zxvf tencent-ailab-embedding-zh-d100-v0.2.0-s.tar.gz

```

Download the bert model and configuration file

```shell

mkdir chinese-bert-wwm-ext

wget -P chinese-bert-wwm-ext https://huggingface.co/hfl/chinese-bert-wwm-ext/resolve/main/added_tokens.json

wget -P chinese-bert-wwm-ext https://huggingface.co/hfl/chinese-bert-wwm-ext/resolve/main/config.json

wget -P chinese-bert-wwm-ext https://huggingface.co/hfl/chinese-bert-wwm-ext/resolve/main/pytorch_model.bin

wget -P chinese-bert-wwm-ext https://huggingface.co/hfl/chinese-bert-wwm-ext/resolve/main/special_tokens_map.json

wget -P chinese-bert-wwm-ext https://huggingface.co/hfl/chinese-bert-wwm-ext/resolve/main/tokenizer.json

wget -P chinese-bert-wwm-ext https://huggingface.co/hfl/chinese-bert-wwm-ext/resolve/main/tokenizer_config.json

wget -P chinese-bert-wwm-ext https://huggingface.co/hfl/chinese-bert-wwm-ext/resolve/main/vocab.txt

```

## Train

```python

python train.py --model_name {model_name} --datadir {datadir} --epochs 30 --lr 1e-4 --max_length 32 --batch_size 8 --savepath ./results --gpu 0 --w2v_path {w2v_path}

```

Or run the scripts

```shell

sh scripts/train.sh

```

## Eval

```python

python eval.py --model_name {model_name} --w2v_path {w2v_path} --model_path {model_path}

```

Or run the scripts

```shell

sh scripts/eval.sh

```

## Inference

```python

python inference.py --model_name {model_name} --batch_size 8 --max_length 32 --savepath ./results --datadir {datadir} --model_path {model_path} --gpu 0 --w2v_path {w2v_path}

```

Or run the scripts

```shell

sh scripts/inference.sh

```

## Results

<div align=center>

| Model | Params(M) | Train Acc(%) |Val Acc(%)|Test Acc(%)|

| :----:| :----: | :----: |:----:|:----:|

| SemNN | 200.04 | 64.02 |65.56|61.41|

| SemLSTM | 200.24 | 66.81 |67.00|69.74|

| SemAttention |200.48| 76.14 |74.50|75.57|

| Bert | 102.27 | 95.85 |82.88|82.65|

</div>