diff --git a/.gitignore b/.gitignore
index abbb866..9a601f6 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,4 +1,4 @@
 __pycache__/
 runs/
-data/
+!data/
 .idea/
\ No newline at end of file
diff --git a/README.md b/README.md
index 150e367..2ed8b3f 100644
--- a/README.md
+++ b/README.md
@@ -1,8 +1,8 @@
 ## Path Ranking with Attention to Type Hierarchies
-This repo contains code for training and testing the proposed models in *Path Ranking with Attention to Type Hierarchies*.
-Due to its large size, data needs to be downloaded separately from [dropbox](https://www.dropbox.com/s/0a4o2jljg4imuux/data.zip?dl=0).
+This repo contains code for AAAI 2020 paper *Path Ranking with Attention to Type Hierarchies*.
+<img src="figure/figure1.png" alt="drawing" width="200"/>
 
-## Notes
+## Resources
 1. Code for baseline models in the paper can be found [here](https://github.com/matt-gardner/pra) (PRA and SFE) and 
 [here](https://github.com/rajarshd/ChainsofReasoning) (Path-RNN).
 2. We provide tokenized data for WN18RR and FB15k-237. Our data format follows 
@@ -23,18 +23,49 @@ FB15k-237 entities can be found [here](https://github.com/thunlp/TKRL).
 3. Install pytorch with cuda
 4. Install requirements by running `pip3 install -r requirements.txt`
 
-## Instruction for running the code
+## Quick Start for Attentive Path Ranking (APR) model
+To help you quickly train and test the model, we have prepared data that are already vectorized. 
+ 
 ### Data
-1. Compressed data file can be downloaded from [dropbox](https://www.dropbox.com/s/0a4o2jljg4imuux/data.zip?dl=0)
+1. Vectorized data file can be downloaded from [dropbox](https://www.dropbox.com/s/0a4o2jljg4imuux/data.zip?dl=0).
 2. Unzip the file in the root directory of this repo.
 
 ### Run the model
-1. Use `run.py` to train and test the model on WN18RR or FB15k-237.
-2. Use `/main/playground/model2/CompositionalVectorSpaceAlgorithm.py` to modify the training settings and hyperparamters.
-3. Use `main/playground/model2/CompositionalVectorSpaceModel.py` to modify the network design. Different attention methods for
+1. Use [`run.py`](run.py) to train and test the model on WN18RR or FB15k-237.
+2. Use [`main/playground/model2/CompositionalVectorSpaceAlgorithm.py`](main/playground/model2/CompositionalVectorSpaceAlgorithm.py) to modify the training settings and hyperparamters.
+3. Use [`main/playground/model2/CompositionalVectorSpaceModel.py`](main/playground/model2/CompositionalVectorSpaceAlgorithm.py) to modify the network design. Different attention methods for
 types and paths can be selected here.
 4. Training progress can be monitored using tensorboardX by running `tensorboard --logdir runs`. Tutorials and Details can be found [here](https://github.com/lanpa/tensorboardX).
 
+## Running the Complete Knowledge Graph Completion Pipeline
+You can also use the code to build knowledge graphs from relation data, sample negative examples, create splits,
+extract paths, vectorize path data, run baselines, and run APR model. 
+
+### Data
+1. Raw data, including relation data, type information, and word embeddings, for WN18RR, FB15k-237 can be downloaded
+from [dropbox]().
+2. Unzip the file in the root directory of this repo.
+
+### Install Dependencies
+1. Clone our [PRA repo](https://github.com/wliu88/pra_scala) and install necessary dependencies. For more details, check
+out readme in that repo.
+2. (Optional) Clone our [CVSM repo](https://github.com/wliu88/ChainsofReasoning) only if you want to run CVSM baselines.
+
+### Run
+1. Use [`demo_wn18rr.py`](demo_wn18rr.py) or [`demo_fb15k237.py`](demo_fb15k237.py) to go through the whole pipeline for the respective dataset.
+2. Use [`run_wn18rr.py`](run_wn18rr.py) or [`run_fb15k237.py`](run_fb15k237.py) if you want to run PRA, SFE, and CVSM baselines
+
+## Code Documentation
+Code are documented with docstrings. We also use `sphinx` to automatically generate API documentation from docstrings.
+
+### Viewing the documentation
+
+1. Make sure you have the necessary `pip` dependencies of `sphinx` installed.
+1. Navigate to [`docs`](docs/) and build the documentation: `make html`
+1. Navigate to [`docs/_build/html`](docs/_build/html) (this folder will be generated by the previous command) and 
+start a simple server: `python -m SimpleHTTPServer`.
+1. Navigate to [`http://localhost:8000`](http://localhost:8000) in your browser.
+
 ## Citation
 Please cite the following work if you find the data/code useful.
 ```text
diff --git a/figure/figure1.png b/figure/figure1.png
new file mode 100644
index 0000000..105f776
Binary files /dev/null and b/figure/figure1.png differ