Skip to content

This is the repo containing code and other resources for the paper entitled "Exploiting Geographical Data to improve Recommender Systems for Business Opportunities in Urban Areas" and published at BRACIS 2019.

License

Notifications You must be signed in to change notification settings

visibilia/SIION

Repository files navigation

Recommender System for discovering of Business Opportunities (Still under construction)

This is the repo containing the code and other resources implemented for the article intitled "Exploiting Geographical Data to improve Recommender Systems for Business Opportunities in Urban Areas", which is published in the proceedings of BRACIS 2019. This framework is part of SIION project, so for ease of communication let's call the content of this repository also as SIION.

Features

SIION is a framework based on clustering and recommender systems to perform some tasks related to Geomarketing. Therefore, the main features are:

  • Geographic Map Partitioning. Given a geographic map corresponding to a specific target city, SIION apply a clustering algorithm to automatically partitioning the target city into a set of non-empty and non-overlapping geographical areas called business districts. To obtain appropriate business districts, the clustering algorithm used should consider different similarity relations among the business locations existing in the target city, e.g. geographical distance, economic factors, etc. In this version, SIION considers the following state-of-the-art clustering algorithms: Kmeans, DBSCAN, agglomerative hierarchical clustering (AC) and business district discovering (BDD).
  • Business Category Recommendation. Given the set of business districts previously discovered, SIION is capable of generate a list of categories recommended to each one of these business districts. In this version, SIION considers the following state-of-the-art recommendation algorithms: Random, most popular (MP), KNN with versions UserKnn and ItemKnn, Group-based recommendation, Poisson Additive Co-Clustering recommendation (PaCo) and Bayesian Personalized Ranking with Matrix Factorization (BPRMF).

Dependencies

In most cases you will want to follow the requirements defined in the requirements/*.txt files in the package.

Base dependencies

scipy
numpy
future
scikit-learn
pandas
matplotlib # for plot some clustering results
python-igraph # for generation of bipartitie graphs which are used in recommendation engines 

How it works?

To be updated

Datasets

To be updated

Visualization

We have work the visualization of our final result using MapBox on the Observable platform. The following figure shows the visualization results obtained for Pittsburgh city using k-means algorithm with k=80, i.e. were computed 80 business districts. Each blue point represents a business location. Business locations inside polygons represent business locations belonging to similar business categories and sharing other similarities (e.g. geographical localization).

alt tag

The complete code for visualization of our results is available here: https://observablehq.com/d/24aea29dedae8bd0

Contributing

This project is open for contributions. Here are some of the ways for you to contribute:

  • Bug reports/fix
  • Features requests
  • Use-case demonstrations

In case you want to implement your own version of this framework, please read more details about our article and SIION project to help you integrate your implementation in our framework.

To make a contribution, just fork this repository, push the changes in your fork, open up an issue, and make a Pull Request!

We are also open to talk more about this project! Just go to our contact form.

Cite

If you used the material available in this repository in your research or project, please cite our work:

@inproceedings{visibilia:siion:2019, 
 author = {Ferreira, Vin\'icius and Valejo, Alan and Valdivia, Paola and Valverde-Rebaza, Jorge},
 title = {Exploiting Geographical Data to improve Recommender Systems for Business Opportunities in Urban Areas},
 booktitle = {Proceedings of The 8th Brazilian Conference on Intelligent Systems},
 series = {BRACIS 2019},
 note = {To be published},
 location = {Salvador-Bahia, Brazil},
 year = {2019}
 }

About

This is the repo containing code and other resources for the paper entitled "Exploiting Geographical Data to improve Recommender Systems for Business Opportunities in Urban Areas" and published at BRACIS 2019.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages