asaldelkhosh / Line-Recommendation

Routes recommendation system in Snappline!

🚖 Line Recommendation System

Implementing a route recommendation system for Snappline! application. In this project we simulate Snappline! routes and search models, after that we create a recommendation system in order to suggest new routes based on users search results on map.

As you may not know, Snappline! is a new taxi application which provides shared cab in specific routes in Tehran. What we are going to do is to simulate the user input data and application data, and build a recommendation system to find out which routes should be add into the system.

Since the stations are constant and we cannot add new stations, we take search results on the map and by using the existing routes on Snappline! we decide which routes should be added into the system.

🦣 HTTP

For simulating the application, we create an http server that generates fake data for us. This http server works like Snappline! app. It stores stations, routes and user search results on map. In order to setup the http server, you can run the following command (make sure to have golang installed on your system):

cd http
go run main.go

request

No you can get all data by making a get request to localhost:8080/data, result is same as the following:

{
  "routes": [
    {
      "id": 1,
      "start": {
        "ID": 10,
        "CreatedAt": "2023-05-09T11:00:12.181975+03:30",
        "UpdatedAt": "2023-05-09T11:00:12.181975+03:30",
        "DeletedAt": null,
        "x": 955,
        "y": -862
      },
      "stop": {
        "ID": 6,
        "CreatedAt": "2023-05-09T11:00:11.2225+03:30",
        "UpdatedAt": "2023-05-09T11:00:11.2225+03:30",
        "DeletedAt": null,
        "x": -584,
        "y": 872
      }
    }
  ],
  "searches": [
    {
      "x1": -545,
      "y1": -850,
      "x2": -215,
      "y2": 471
    }
  ]
}

🤖 Data

Out model uses pandas library in order to read the json data and convert it into a dataframe. After that we use ast in order to convert the inner json objects into a directory. In all first steps, we are trying to make our data clean and useable for processing. Our pseudo code is as follow:

[1] http request to get json data.
[2] convert json data to `pandas` dataframe.
[3] create `routes` and `search` dataframe.
[4] map the `search` entities to a station/node in `routes` dataframe. (since we cannot create stations)
[5] map the `search` source and destination to a new route in `routes` dataframe.
[6] count the number of new entities.
[7] sort them in decreasing order.

After we cleaned our data we create two dataframes for search and routes. We convert each entity to shapely.geometry.Point object to use the nearest_points method in order to apply our algorithm on the data to map them into routes dataframe. Now we perform nearest_points to map a search to route and then we use value_counts method in pandas dataframe in order to get the results.

execute

You can execute the script by running python main.py but make sure to have all of the libraries in requirements.txt installed on an virtual environment. Result would be as follow:

POINT (805 683)-POINT (-408 834)      10
POINT (805 683)-POINT (-629 -733)     6
POINT (805 683)-POINT (475 -265)      4
                                     ..
POINT (-660 562)-POINT (-666 -109)    1
POINT (-245 -149)-POINT (805 683)     1