ARMUT — Association Rule Learning Project

Business Problem

Armut, Turkey’s largest online service platform, brings together service providers and those seeking services. It facilitates easy access to services like cleaning, renovation, and transportation with just a few taps on a computer or smartphone.

Get Yasemin Derya Dilli’s stories in your inbox

Join Medium for free to get updates from this writer.

Remember me for faster sign in

By utilizing the dataset containing users who have received services, along with the services they have availed and the respective categories, the objective is to create a product recommendation system using Association Rule Learning.

UserId: Customer ID
ServiceId: Anonymized services belonging to each category. (Example: Under the Cleaning category, there might be a service for sofa cleaning.)
A ServiceId can be found under different categories and represents different services under different categories.
(Example: ServiceId 4 with CategoryId 7 represents a service for radiator cleaning, while ServiceId 4 with CategoryId 2 represents a service for furniture assembly.)
CategoryId: Anonymized categories. (Example: Cleaning, transportation, renovation categories)
CreateDate: The date when the service was purchased

First of all we import libraries and read dataset

import pandas as pd
pd.set_option('display.max_columns', None)
from mlxtend.frequent_patterns import apriori, association_rules

df_ = pd.read_csv("/kaggle/input/armut-dataset/armut_data.csv")
df = df_.copy()
df.head()

df["Hizmet"] = [str(row[1]) + "_" + str(row[2]) for row in df.values]
df.head()

We have created a new variable representing services by combining ServiceID and CategoryID with ‘_’. ServiceID encompasses CategoryID.

df["CreateDate"] = pd.to_datetime(df["CreateDate"])
df.head()
df["NEW_DATE"] = df["CreateDate"].dt.strftime("%Y-%m")
df.head()
df["SepetID"] = [str(row[0]) + "_" + str(row[5]) for row in df.values]
df.head()

To apply Association Rule Learning, it is necessary to create a basket (invoice, etc.). Therefore, we derived a new basket variable from the date variable we have.

df[df["UserId"] == 7256 ]

We are viewing all the services purchased by the user with UserId 7256.

invoice_product_df = df.groupby(['SepetID','Hizmet'])['Hizmet'].count().unstack().fillna(0).applymap(lambda x: 1 if x > 0 else 0)
invoice_product_df.head()

We have created a pivot table of service baskets as shown above.

frequent_itemsets = apriori(invoice_product_df, min_support=0.01, use_colnames=True)
rules = association_rules(frequent_itemsets, metric="support", min_threshold=0.01)
rules. Head()

We have generated association rules using the Apriori algorithm.

def arl_recommender(rules_df, product_id, rec_count=1):
    sorted_rules = rules_df.sort_values("lift", ascending=False)
   
    recommendation_list = [] 

    for i, product in sorted_rules["antecedents"].items():
        for j in list(product): 
            if j == product_id:
                recommendation_list.append(list(sorted_rules.iloc[i]["consequents"]))

    recommendation_list = list({item for item_list in recommendation_list for item in item_list})
    return recommendation_list[:rec_count] 



arl_recommender(rules,"2_0", 4)

['13_11', '22_0', '38_4']

We have utilized the ‘arl_recommender’ function to recommend a service to a user who last availed the service 2_0.

Big thanks to Vahit Keskin and Miuul

Contact me on Linkedin :) yaseminderyadilli