Data Analyst Approach to Get Demand Products based on Seasonality for e-Commerce
In Indonesia, there are many seasons that are very profitable for business people, especially the e-commerce business. For example, there are lots of users who will buy Christmas products when it’s time for Christmas, lots of sales for raincoats and sweaters when it is close to the rainy season, kurma and other Islamic products sell very high when it comes to Ramadan and Eid. This makes it a business opportunity, especially to offer products that are seasonal according to the season.
Being at the forefront of displaying and offering superior products is not easy for me as a Data Analyst product recommendation. If the product that is displayed is not suitable, the user will feel that the recommendations given are not personal. For the rest, although appropriate, there is a possibility that the timing may not be correct.
Therefore, in this article, I will explain a little about my project regarding seasonal products based on the month ratio.
The point of seasonal products are highly desirable in certain months but less desirable in other months - (Schafer et al. 2001)
From Schafer’s quote, an idea emerged to calculate the month ratio, by comparing the number of sales months to N, months N-1, and months N+1 to total sales in one year.
It should be remembered because if we only use sales in one year, the data used is less, therefore, we can use the last 3 years of data to predict whether the product is a seasonal product in a certain month or not. Before calculating the month ratio, it is better if we first calculate standardized sales (because it uses 3 years of data) so that each product will have a standardized sales value every month. The illustration can be seen in the table below:
Step 1 : Prepare monthly sales per products
Step 2 : Standadized monthly sales and calculate final standardize values per month
To calculate the standardized value, you can use the formula:
Furthermore, from the standardized sales value, we can calculate the month ratio for a product, which can be calculated using the formula:
where pi denotes product i, and the denominator is the sum of all orders for the product i within the last twelve months (in the 3 years). This calculation is done for each item product and each month. A product is said to be seasonal if the product has a month ratio greater than the threshold that we specify (in this project I use a threshold of 0.5 with the assumption that one product only has a maximum of two seasonal months in one year).
From the calculation of the month ratio, you can find the number of seasonal products each month in the chart below:
It is found that the early months (January to April) have the most seasonal products when compared to other months.
From the results of this month ratio calculation, to ascertain whether the model used is better or not, the precision value will be calculated every month by calculating:
And from this model, it is found that the precision product-seasonal products are higher than all products. Where from these results it can be concluded that displaying seasonal products, will reduce the recommendation slot but the precision obtained is much higher than if we randomly display the products.
References :
- Eka Wijaya, G., and Yong-han, L. (2012). Hybrid Recommender System with Seasonality. Asia Pacific Industrial Engineering and Management Systems Conference (https://apiems.org/).
- Stormer, Henrik. (2007). Improving e-commerce recommender systems by the identification of seasonal products. Association for the Advancement of Artificial Intelligence (www.aaai.org).
- Khullar, K., 2018. Understanding Ecommerce Seasonality And Identifying Niche Based Seasonal Events. [online] https://www.semrush.com/. Available at: <https://www.semrush.com/blog/understanding-ecommerce-seasonality-identifying-niche-seasonal-events/> [Accessed 26 November 2020]
Collaboration Project with Aulia Muthia and jevi fronatalia
